Real-time medical images enhancement technique using Adaptive Histogram Equalization model.

Real-time medical images enhancement technique using Adaptive Histogram Equalization model - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Sci Rep . 2026 Apr 10;16:16928. doi: 10.1038/s41598-026-45697-6 Real-time medical images enhancement technique using Adaptive Histogram Equalization model Arvind Dhaka 1 , Arpit Kumar Sharma 1, ✉ , Amita Nandal 2, ✉ Author information Article notes Copyright and License information 1 Department of Computer and Communication Engineering, Manipal University Jaipur, Jaipur, India 2 Department of IoT and Intelligent Systems, Manipal University Jaipur, Jaipur, India ✉ Corresponding author. Received 2025 Oct 8; Accepted 2026 Mar 20; Collection date 2026. © The Author(s) 2026 Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ . PMC Copyright notice PMCID: PMC13230594  PMID: 41963405 Abstract In low-contrast modalities like magnetic resonance imaging, medical image augmentation helps reveal anatomical structures and diseased locations. This paper presents an adaptive histogram-based contrast enhancement technique to improve medical picture quality while conserving structure. The suggested approach enhances tumor areas and tissue borders without artifacts using statistical normalization, adaptive histogram modeling, Gaussian-based intensity redistribution, and brightness-preserving contrast modification. T1, T2, FLAIR, and contrast-enhanced T1 MRI scans from the BraTS 2020 Brain Tumor Segmentation Dataset were used in experiments. The proposed method was compared to five state-of-the-art contrast enhancement methods: Histogram Equalization (HE), Contrast Limited Adaptive Histogram Equalization (CLAHE), Gamma Correction (GC), Retinex-based enhancement, and Adaptive Gamma Correction. Quantitative metrics like PSNR, MSE, SSIM, and entropy were used to evaluate the method’s performance. The suggested technique outperforms HE (31.46 dB), CLAHE (34.15 dB), Gamma Correction (32.74 dB), Retinex (35.21 dB), and Adaptive Gamma Correction (36.04 dB) in experimental PSNR (37.82 dB). The suggested technique has an SSIM of 0.962, compared to 0.891, 0.915, 0.902, 0.934, and 0.947 for the competing approaches. The approach decreases reconstruction error with an MSE of 18.4, lower than HE (42.7) and CLAHE (29.6). Qualitative analysis shows better contrast and tumor borders without brightness amplification. These findings show that the adaptive histogram approach enhances medical image analysis and diagnostic assistance systems robustly and efficiently. Keywords: Image enhancement, Image processing, Grey-scale, Image segmentation, Histogram, Medical image, MRI Subject terms: Computational biology and bioinformatics, Engineering, Health care, Mathematics and computing, Medical research Introduction The quality of medical Image (MI) has turned into a fundamental portion of current medication and straightforwardly impacts the exactness of specialists’ treatment as well as diagnosis. Low contrast and resolution in MIs have made right diagnosis troublesome; this straightforwardly impacts the accuracy and speed of specialists’ prognosis. Hence, it becomes essential to improve medical picture enhancement to reflect the data of a disease more precisely and clearly 1 , 2 , 3 . Medical imaging (MI) 1 uses several innovations to create pictures of the human body’s internal functions/structures 2 . Healthcare experts (HEs) 3 utilize these clinical pictures for four purposes: diagnosis 4 , treatment research, monitoring 5 and planning 6 . The HEs, first and foremost, use clinical pictures to recognize and analyze ailments that might be convoluted or difficult to detect without imaging. Besides, clinical images assist HEs with plan and get ready for medical processes. For instance, computed tomography (CT) 7 , 8 filters help HEs in arranging radiation treatment for cancer patients by initiating the recognizable site and size of the cancer areas. Third, clinical images have been utilized to screen the growth of infections after some time. At last, researchers can utilize clinical pictures to concentrate on the physiology/anatomy of the human body 9 and explore the impacts of ailment and therapies on the body 10 . Fractional differentials that becomes a hypothesis of arbitrary order, are like integral differentials 4 , 5 . Compared to conventional integral differential methodologies, fractional differentials applicable to image processing can improve edges, make texture details more clear, and safeguard smooth regions 6 , 7 , 8 . Consequently, MIs processed by partial differentials have greater contrast and are more clear. Conventional partial differentials utilize a similar fragmentary order to deal with smooth image regions, textures and edges; notwithstanding, while edges would be upgraded by great fractional orders, smooth regions and weak textures would be overlooked, and keeping in mind that more fragile surfaces and smoother regions would be protected by low fractional orders, edges would be debilitated. Hence, picture enhancement is hard to accomplish in practice. To deal with such problems, conventional and enhanced on fractional differential programming have been created for digital image processing 9 , 10 , 11 , 12 . Adaptive fractional derivatives for picture denoising issues have additionally been considered 13 , 14 , 15 . MI is currently becoming fundamental in various biomedical examination and clinical practice domains. Researcher concentrate on cells and create 3D confocal microscopy data sets 11 . Neuro-researchers distinguish local metabolic mind action from positron emission tomography (PET) 12 , 13 , magnetic resonance spectrum imaging (MRSI) 14 and functional magnetic resonance imaging (fMRI) scans. Virologists produce 3D reproductions of infections from micrographs 15 . Radiologists can distinguish and measure growths from X-ray and CT examines 16 . Moreover, MI incorporates a few different strategies: X-ray scans, nuclear medicine imaging (NMI) and ultrasounds and so forth. X-rays to make pictures of bones. Ultra-sounds utilize the sound waves to assemble pictures of internal organs. The NMI 17 utilizes a modest quantity of radioactive material to make pictures of the body’s functions and interior structures, for example, blood stream, digestion, and organ capability. Then again, medical image processing (MIP) stays an issue for specialists. MIP generally contains seven general operations, as displayed in Fig. 1 . Image acquisition includes procuring clinical pictures from different MI modalities, for example, ultrasound scans, MRI scans, CT scans and X-rays. Preprocessing 18 assists with up surging and cleaning the clinical pictures by enhancing contrast, correcting distortion and eliminating noise. Segmentation 19 isolates the region of interest from the foundation and isolates various zones in the picture. Registration 20 adjusts different clinical pictures to make a composite picture that gives a more extensive perspective on the pathology/anatomy. Feature extraction separates significant elements from clinical pictures, like different structures’ size, intensity, texture and shape. Classification characterizes various configurations inside clinical pictures in view of their properties/features. Visualization 21 assembles visual portrayals of clinical pictures in 2D or 3D to help research, treatment planning and diagnosis. Fig. 1. Open in a new tab Algorithmic flow chart for the proposed method. MIP, first and foremost, can be utilized to decide the vasculature, volume and diameter of an organ/tumor 22 , blood flow parameters 23 or different liquids, and microscopic variations that still can’t seem to raise any generally noticeable flags. Besides, MIP makes practical human body models/simulations that are utilized for clinical schooling 24 as well as training purposes 25 . Thirdly, MIP gives real-time visualization and guidance during surgeries 26 , permitting doctors to really explore complicated physical structures. At long last, MIP helps in making custom prosthetic gadgets 27 , for example, orthotics/implants 28 that are custom-made to the patient’s requirements and anatomy. Related work The majority of the current improvement strategies are experimental or heuristic techniques — emphatically connected with a specific images type — that for the most part target further developing the contrast level of pictures reduced at the time of acquisition procedure 29 . Truly, searching the best gray level mapping that adaptively upgrades every unique input picture can be viewed as an optimization issue 30 . Tragically, no unifying theory utilizing a normalized picture quality measure is presently accessible to characterize a common standard for picture improvement 31 . Likewise, on account of MI, strategies custom-made on unambiguous operations are important to accomplish a critical enhancement and, commonly, intuitive methodology including significant human exertion are expected to get good outcomes. To accomplish such goals as well as reproducible estimations passing on medically valuable data, administrator reliance ought to be limited through computerized techniques. Point-wise activities in the spatial (pixel) space, addressing the easiest type of image processing, are successful arrangements since proficiency necessities have likewise to be met. On account of image improvement, they re-map every input gray level into a specific resultant gray level, as indicated by a worldwide transformation 32 . Consequently, these sort of methods treat pictures all in all, disregarding unambiguous features of various districts, or specifically differing amid the collection of contrast enhancement settings/degrees 31 . Histogram Equalization (HE) is the most well-known worldwide picture enhancement strategy, whose point is to consistently redistribute the input gray level qualities as per the cumulative density function (CDF) of its histogram 32 . Unfortunately, HE doesn’t consider the image mean intensity 33 i.e. dependent upon a tremendous variation at the time of equalization procedure by perpetually moving the result mean brightness to the center gray level, no matter what the mean gray level in the input picture 34 . Thus, HE cannot safeguard the input mean brightness, potentially experiencing the over-enhancement, and bringing about the artifacts like the alleged washed out impact 33 . This worldwide change strategy applies contrast extending simply on gray levels with the greatest frequencies, creating a critical contrast misfortune concerning the gray levels classified by low frequencies in the information histogram 35 . Bi-Histogram Equalization (Bi-HE) i.e. upgraded variant of the conventional HE, was suggested to conquer the restrictions connected with input mean brightness preservation, mostly created by histogram flattening 35 . Primarily, Bi-HE divides the original histogram into two sub-histograms as indicated by the global mean of the first picture; further, the sub-histograms are freely handled by applying the standard HE strategy to every one of them. The intricacy of the enhancement criteria to be fulfilled (such as., the efficient contrast stretching joined with picture detail preserving) prompts the application of worldwide pursuit meta-heuristics that take into consideration adapting to a few imperatives, which are not commonly manageable through customary thorough computational methodologies 31 , 32 , 33 . Evolutionary techniques have been generally embraced in the image improvement space to track down the ideal enhancement kernal 31 , filter sequences 36 , or transformation of input-output transformation. In recent time, Hashemi, Kiani, Noroozi, and Moghaddam suggested a GA-based technique that effectively encodes the histogram through the non-zero intensity levels, by implementing genetic administrators that straightforwardly process images to enhance the noticeable details as well as contrast of lesser illumination areas, particularly on account of great dynamic ranges 37 . The researchers contended that this strategy yields “normal looking” pictures considering the visible appearance. Advanced signal processing and machine learning have improved picture quality, feature representation, and automated analysis in medical and computational imaging. In ultrasound imaging, deep learning has improved resolution and visualization, allowing accurate reconstruction of microvascular structures and improved diagnostics 38 . Adaptive graph-based anomaly detection frameworks for hyperspectral images capture complicated spectral–spatial interactions and identify aberrant patterns better 39 . Perceptual information fidelity models connect computer evaluation with human perception to determine picture quality more correctly 40 . Several research have examined sophisticated biological and medical dataset analytics technologies. The use of iterative adaptive mini-minimum spanning tree techniques to identify outliers in medical datasets improves data dependability for downstream analysis 41 . Deep generative networks have been used to estimate disparity in ste-reo-endoscopic images to enhance depth perception and assist minimally invasive surgery 42 . New contour tracking methods based on time-varying internal model principles have enhanced dynamic visual tracking and medical imaging accuracy 43 . Recent medical image analysis research emphasizes multimodal learning and noise reduction. Dermatological illness classification using multimodal masked autoencoder frameworks with adjustable masking improves feature representation and classification accuracy in difficult clinical imaging datasets 44 . Adaptive wavelet thresholding and complementing ensemble empirical mode decomposition may help reduce imaging data noise while keeping signal characteristics 45 . Advanced encryption solutions like dynamic Hill ciphers for medical Internet-of-Things contexts address medical imaging system security and privacy 46 , 47 . Shehu Aliyu et al. proposed a chaotic salp swarm algorithm for feature selection in imbalanced microarray cancer gene classification, showing that metaheuristic optimization can identify informative genes and improve classification performance in high-dimensional biomedical datasets 48 . According to Karimi et al.‘s comprehensive survey, feature selection techniques in big medical databases are becoming more important because dimensionality reduction methods improve computational efficiency and predictive accuracy in large-scale healthcare data analytics 49 . Yousaf et al. demonstrated how mobile health (m-health) systems and intelligent monitoring and assistive technologies might enhance dementia and Alzheimer’s disease patient care and remote medical supervision 50 . Image analysis and feature selection are well-studied in dermatology. Comparative study by Javed et al. demonstrated that discriminative visual qualities improve automated skin cancer detection methods for dermoscopic skin lesion identification 51 . Recent study combines secure medical data management with strong machine learning. A CNN-enabled skin cancer detection system with blockchain-based cloud storage by Nasir et al. secures medical data access and classification accuracy in scattered healthcare settings 52 . Khan et al. improved dermatological image analysis diagnosis reliability by unifying skin lesion segmentation and classification using deep neural network feature fusion and selection 53 . In their study of fundus image classification methods for glaucoma detection, Saba et al. stressed the importance of feature extraction and classification 54 . These studies indicate that feature selection, deep learning, and secure data management frameworks increase intelligent healthcare system accuracy, efficiency, and reliability. Recent research show that picture preprocessing and enhancement improve automated image analysis systems in medical and security imaging. The research shows that edge augmentation and frequency-domain changes improve convolutional neural network discriminative features, enabling more accurate recognition of complex objects in crowded pictures 55 . The authors also note that CNN-based models and augmentation methods increase detection resilience in demanding visual conditions. The proposed method reduces contrast stretching due to artifact control. Augmentation cannot change diagnostic data or create fake edges. To balance contrast enhancement and structural integrity, the transformation function is tuned to the cumulative intensity distribution in medical imaging, where tiny characteristics might change diagnostic results. Unlike to visual enhancing approaches, the suggested solution maximizes quantitative performance using objective measurements. PSNR, MSE, and structural preservation increase concurrently with signal-to-noise and structural similarity improving. The suggested histogram equalization approach differs from contrast distribution-optimizing methods by its multi-objective improvement. Methodology In the proposed method, the color image is first switched into completely to grayscale and afterward computationally handled to get the probability density functions (pdf) and cdf values for grayscale HE. At last, the contrast among the first grayscale picture and the equalized picture is utilized to initiate the hue preservation as well as acquire an equalized color picture. The investigation included utilizing retina, brain and knee bone pictures to evaluate the efficiency of several picture contrast improvement techniques. Image histogram is the premise programming for various spatial domain processing. It offers valuable picture statistics as well as it becomes beneficial for picture segmentation, compression and enhancement. It becomes easy to measure in programming as well as in hardware execution. HE creates a picture whose levels of intensity are similarly possible and covers the whole picture 2 . The consequence of this intensity level equalization becomes a picture with expanded dynamic range prompting contrast increment 1 . There are various histogram types HE programming like Quadrant Dynamic Histogram Equalization (QDHE), Contrast-Limited Adaptive Histogram Equalization (CLAHE) and Cumulative Histogram Equalization (CHE). Through a multi-stage processing pipeline, the proposed architecture improves medical picture contrast while conserving clinically essential structural information. Preprocessing, histogram analysis, adaptive transformation, contrast redistribution, and image reconstruction comprise the architecture (Fig. 1 ). Medical pictures from modalities including MRI, CT, and X-ray are adjusted using mean-variance normalization to eliminate scanner-related intensity fluctuations. In the preprocessing stage, the input image of size undergoes normalization and noise reduction to ensure uniform intensity distribution across the dataset. The normalized intensity is expressed as: 1 Instead of histogram equalization, the suggested technique uses adaptive histogram transformation and Gaussian-based intensity redistribution to balance contrast enhancement and avoid severe intensity overamplification. An adaptive updating system dynamically modifies histogram bins depending on local intensity characteristics and brightness limitations to increase contrast without noise distortions. For the final increased output, inverse transformation and intensity restoration reconstitute the improved picture. The suggested technique is assessed using quantitative measures including MSE, PSNR, and SSIM, which quantify reconstruction accuracy, signal quality, and structure preservation in improved medical pictures. Model structure, parameter choices, and processing pipeline Model structure The proposed model uses a multi-stage enhancement architecture to increase medical picture contrast while keeping clinically relevant structural information. The framework consists of five major modules: preprocessing, histogram modeling, adaptive contrast transformation, histogram updating, and image reconstruction. The recommended implementation strategy pseudo-code is shown in Fig. 2 . Input and Preprocessing Module Fig. 2. Open in a new tab Pseudo-code for the proposed implementation strategy. The system receives a medical image obtained from imaging modalities such as Magnetic Resonance Imaging, Computed Tomography, or X-ray Radiography. If the image is in RGB format, it is converted into grayscale intensity values: 2 Normalization is applied to standardize intensity distribution: 3 where and denote the mean and standard deviation of pixel intensities. Optional filtering (median or Gaussian) is applied to reduce noise while preserving edges. (2) Histogram Modeling Module The algorithm models the intensity distribution using histogram statistics. For an image with intensity levels: 4 where represents the number of pixels with intensity level . The normalized histogram (probability density function) is computed as: 5 The cumulative distribution function (CDF) is calculated as: 6 These statistics guide the adaptive contrast transformation. (3) Adaptive Histogram Transformation Module Instead of applying standard histogram equalization, the proposed method uses adaptive histogram redistribution. The transformation function is defined as: 7 To stabilize the enhancement process, the intensity distribution is mapped to a Gaussian distribution: 8 This step avoids extreme intensity stretching and reduces artifact formation. (4) Histogram Updating and Contrast Control The histogram bins are dynamically adjusted using an adaptive weighting parameter: 9 where is adaptive enhancement factor, is uniform histogram component and is locally estimated histogram value. Brightness constraints and ensure controlled contrast stretching: 10 This mechanism prevents over-enhancement and preserves important anatomical structures. (5) Image Reconstruction Module After histogram transformation, the enhanced image is reconstructed. If transform-domain processing is used, the inverse transformation is applied: 11 Finally, exponential restoration is applied: 12 The resulting image exhibits improved contrast and structural clarity. Parameter choices The performance of the proposed model depends on several parameters controlling enhancement strength and noise suppression. The parameters in Table 1 ensure balanced enhancement, noise suppression, and artifact prevention. Table 1. List of parameters. Parameter Description Typical value Number of gray levels 256 Mean intensity Dataset dependent Standard deviation Dataset dependent Adaptive histogram weight 0.4–0.7 Contrast control factor 0.3–0.5 Lower brightness threshold 0.1–0.2 Upper brightness threshold 0.8–0.9 Filter size Noise reduction kernel 3 × 3 or 5 ×  5 Open in a new tab Histogram equalization (HE) Histogram equalization (HE) may cause over-enhancement, noise amplification, and loss of local structural features in medical imaging, when pixel intensity changes correlate to intricate anatomical structures. The proposed system uses algorithmic algorithms and assessment mechanisms to reduce artifact creation and preserve diagnostic picture quality. Adaptive histogram equalization with controlled contrast enhancement minimizes intensity value stretching. In traditional histogram equalization, the transformation function redistributes gray levels throughout the intensity range using the cumulative distribution function (CDF). The CDF transformation can be expressed as. 13 where represents the input gray level, is the probability density function (PDF), and is the number of gray levels. While this global transformation improves contrast, it may amplify noise or produce unnatural brightness shifts in homogeneous regions. To mitigate this issue, the proposed method applies localized histogram processing, where the image is divided into smaller contextual regions and the transformation is computed locally: 14 where represents the local histogram distribution around pixel location . This localized enhancement preserves fine anatomical details and prevents the global redistribution from distorting specific regions. The algorithm incorporates a contrast-limiting mechanism that constrains the amplification of histogram bins. When the frequency of a particular gray level exceeds a predefined threshold , the histogram values are clipped: 15 The clipped values are then redistributed uniformly across the histogram bins: 16 This process prevents excessive contrast amplification and reduces the likelihood of artificial edges or halo artifacts. Enhancement incorporates brightness preservation limitations to maintain local image structures. The mean intensity of the enhanced image is maintained close to the mean intensity of the original image : 17a 17b where is a small tolerance value. This constraint helps avoid excessive brightness shifts that could produce unrealistic visual appearances. Redistribution pixel intensity values The proposed lgorithm analyzes the intensity distribution of the grayscale image , where intensity values and represents the number of gray levels. The histogram counts the number of pixels at each intensity level, and the normalized histogram represents the probability density function of gray levels for an image of size . The cumulative distribution function is then used to transform pixel intensities through the mapping , which redistributes intensity values across the available dynamic range. Because this transformation is monotonic, the relative ordering of pixel intensities remains unchanged, ensuring that structural features such as edges and anatomical boundaries are preserved while contrast is enhanced. To avoid the over-enhancement associated with classical global histogram equalization, the proposed method employs adaptive local histogram processing within a neighborhood window . The local histogram probability and corresponding transformation allow the algorithm to enhance contrast specifically in regions with limited intensity variation, which is common in soft tissues of medical images. A contrast-limiting mechanism is also applied by clipping histogram values that exceed a threshold , followed by redistribution of excess counts across all intensity bins. This step prevents excessive amplification of certain gray levels and reduces the risk of artifacts. Additionally, brightness preservation is maintained by ensuring that the mean intensity of the enhanced image remains close to the original mean intensity, i.e., , thereby preventing unrealistic brightness shifts that could misrepresent tissue characteristics. To further stabilize the enhancement process, preprocessing may include frequency-domain operations such as the Fourier transform and logarithmic magnitude scaling , which suppress extreme frequency components and reduce noise influence. After processing, the inverse transform reconstructs the enhanced image while preserving phase information essential for maintaining spatial structures and edges. The final enhanced image can therefore be expressed as , where represents noise suppression and denotes the adaptive histogram transformation. Improved visual clarity and diagnostic visibility The work improves CT scans and X-rays’ visual clarity and diagnostic visibility using histogram-based contrast enhancement approaches. Histogram Equalization (HE), Adaptive Histogram Equalization (AHE), and Contrast-Limited Adaptive Histogram Equalization (CLAHE) improve picture contrast by shifting pixel intensity values. Initial processing includes adaptive histogram augmentation of the medical picture. Input the original picture and initialize the settings. Color images are transformed to grayscale to ease intensity analysis. The PDF and CDF of pixel intensities are then calculated to define gray level distribution and decrease noise. Next, picture processing is done using frequency-domain methods such the DCT or Fourier Transform to examine frequency component magnitude. Estimating magnitude coefficients via logarithmic scaling compresses large dynamic range data and emphasizes low-intensity structures. The intensity distribution throughout the picture is determined by measuring the histogram coefficients. A Gaussian distribution model is constructed from this distribution to guide histogram mapping. Gaussian-based histogram shifting and exponential mapping change pixel intensities to increase contrast while retaining structural integrity. The resulting picture has better contrast and decreased noise after an inverse transform recovers phase information and reconstructs the upgraded image. In the Histogram Equalization (HE) stage, the grayscale image intensity distribution is normalized using the probability density function defined as , where represents the number of pixels with intensity level , is the total number of pixels, and is the total number of gray levels. The cumulative distribution function is then calculated as . This cumulative function serves as the transformation mapping that redistributes intensity values over the entire gray-level range, thereby improving the global contrast of the image. In color or multi-channel image equalization, hue preservation eliminates color distortion. This technique generates an equalized color picture by retaining the ratio between the original grayscale image and the equalized image, and then reconstructs a hue-consistent color image using the difference. These approaches ensure augmentation does not significantly change medical image color. MSE and PSNR measure enhancement quality. PSNR assesses perceived reconstruction quality and contrast improvement, whereas MSE measures the average squared difference between original and upgraded images. Better enhancement with less distortion typically means greater PSNR. AHE boosts contrast by localizing histogram equalization. Histogram equalization is done individually after contextualizing the image. Local augmentation clarifies tiny structures and low-contrast areas in medical imaging. The histogram of an image with possible intensity levels is defined as , and the normalized histogram is , where and represent the image dimensions. The transformation function is derived from the cumulative distribution function: . This mapping redistributes pixel intensities across the available gray-level range to improve local contrast. The threshold-based Adaptive Histogram Equalization approach normalizes and updates histograms. The method examines picture pixels, computes intensity variations between nearby pixels, and updates histogram bins when intensity variation exceeds a threshold. Normalization adjusts enhancement strength. A contrast control factor scales bins outside the brightness range, while a weighted mix of local histogram statistics and global normalization settings updates bins inside it. Contrast is increased without noise via adaptive weighting. Noise handling Handling noise in medical images is essential before applying histogram-based contrast enhancement because noise can distort the gray-level distribution and degrade diagnostic information. In the proposed framework, different noise models—Gaussian noise, salt-and-pepper noise, and speckle noise—are addressed through mathematical preprocessing operations followed by histogram-based enhancement. The general degraded image model can be expressed as , where is the original noise-free image, is the observed noisy image, and represents the noise component. The goal of preprocessing is to estimate from before applying histogram equalization techniques. Gaussian noise handling Gaussian noise is an additive noise that follows a normal probability distribution. It is commonly produced by sensor electronics during image acquisition. The noisy image model is expressed as , where the Gaussian noise term follows with mean and variance . The probability density function of Gaussian noise is. To reduce Gaussian noise, a Gaussian smoothing filter is applied where is the Gaussian kernel. This filtering operation attenuates high-frequency noise components while preserving the main anatomical structures of the image. After denoising, histogram-based contrast enhancement can be applied safely. Salt-and-pepper noise handling Salt-and-pepper noise is an impulse noise where pixels are randomly replaced by minimum or maximum intensity values (0 or 255 in an 8-bit image). The noise model can be defined as where and represent the probabilities of salt and pepper noise respectively. To remove impulse noise, a median filter is commonly used where denotes a neighborhood window around pixel . The median operation effectively removes isolated noisy pixels while preserving edges and fine structures, which is particularly important for medical images. Speckle noise handling Speckle noise is a multiplicative noise that frequently appears in coherent imaging systems such as ultrasound or radar-based imaging. The noisy image can be modeled as where represents multiplicative speckle noise. To convert multiplicative noise into additive form, a logarithmic transformation is applied . Let hen where behaves like additive noise. Conventional denoising filters (Gaussian or median filters) can then be applied. After filtering, the enhanced image is reconstructed using the exponential transformation . Histogram-based enhancement after noise reduction After noise suppression, contrast enhancement is performed using histogram-based techniques. The histogram of an image with gray levels is where is the number of pixels with intensity level . The normalized histogram (probability density function) is where is the image size. The cumulative distribution function (CDF) is The histogram equalization transformation function is which redistributes the gray levels across the available intensity range to enhance contrast. For adaptive enhancement, the transformation is applied locally within subregions of the image where is the local histogram within a neighborhood region centered at . Generalization to different noise types Histogram-based enhancement methods such as HE, AHE, and CLAHE are not inherently noise-specific. They operate on the statistical distribution of pixel intensities rather than assuming a particular noise model. However, if applied directly to noisy images, these techniques may amplify noise because noise alters the histogram distribution. To address this issue, the proposed framework integrates noise-adaptive preprocessing before histogram transformation. Let where represents a noise-dependent filtering function (Gaussian, median, or logarithmic filtering). Histogram enhancement is then applied as Thus the overall enhancement model becomes . This two-stage process ensures that histogram equalization operates on a noise-reduced image, allowing the method to generalize across multiple noise types. Dataset and experimental details This work uses an automated brain tumor segmentation dataset, BraTS 2020, is used [ https://www.kaggle.com/datasets/awsaf49/brats2020-training-data ]. T1-weighted, T2-weighted, FLAIR, and contrast-enhanced T1 (T1ce) MRI scans can characterize tumors. T1ce highlights tumor boundaries and active tumor regions following contrast injection, whereas T1 shows normal anatomy. FLAIR improves tumor-related edema imaging by decreasing cerebrospinal fluid signals. T2 images indicate edema and abnormal fluid accumulation. About 369 annotated patient cases are available as three-dimensional MRI volumes with a mean resolution of 240 × 240 × 155 voxels. This study used expert-annotated tumor segmentation mask pictures from the training subset. To ensure consistent analysis and multi-modal learning, only whole patient cases with all four MRI modalities were included. Strong image enhancement and segmentation methods may be tested using clinical data from many institutions and imaging modalities. Pretreatment and standardization before analysis improved data consistency and image quality. MRI volumes were aligned around a common template and resampled to a 1 × 1 × 1 mm³ isotropic voxel resolution. After skull stripping eliminated non-brain tissues and background artifacts, the model concentrated on key brain areas. MRI scanner and acquisition setup variability was decreased by z-score intensity normalization. Denoising, contrast enhancement, and histogram normalisation showed tumor boundaries and enhanced signal-to-noise ratio. Cropped or divided image volumes (128 × 128 × 128) surrounding brain or tumor regions were used to reduce computation and improve model training efficiency. Evaluation metrics and computational efficiency The proposed histogram-based enhancement technique was assessed using Peak Signal-to-Noise Ratio (PSNR), Mean Squared Error (MSE), and Standard Deviation (SD), which are commonly used quantitative measures for medical imaging picture quality and contrast improvement. These measures complement reconstruction fidelity, noise reduction, and intensity distribution, which are crucial for clinical imaging enhancement algorithm evaluation. The Mean Squared Error (MSE) measures the average squared difference between the original image and the enhanced image . It quantifies the overall distortion introduced during the enhancement process and is defined as , where and denote the image dimensions. The enhancement approach maintains picture information with minimum distortion if MSE levels are low. This is crucial in medical pictures since significant distortion might affect anatomy or diagnostics. The Peak Signal-to-Noise Ratio (PSNR) measures the ratio between the maximum possible intensity value and the distortion represented by MSE. It is expressed as where is the maximum possible pixel value (typically 255 for 8-bit images). PSNR increases reconstruction quality and reduces deterioration. In medical imaging, high PSNR indicates that the improved picture retains a high degree of similarity while improving contrast. The Standard Deviation (SD) of pixel intensities reflects the spread or variability of gray-level values within an image and is often used as a measure of image contrast. It is calculated as , where is the mean intensity of the image. Higher SD values suggest a broader intensity dispersion and better contrast. Increased contrast helps medical image enhancement expose cancers, lesions, and tissue borders that are faint in low-contrast pictures. These three measurements give quantitative insights, but they focus on pixel-level differences and statistical contrast, which may not completely represent clinical perceptual or structural quality. To address this, new assessment criteria were proposed. A typical measure is the Structural Similarity Index (SSIM), which compares pictures based on brightness, contrast, and structure which is written as where and represent the mean intensities, and represent variances, and represents the covariance between the original and enhanced images. The constants and stabilize the computation. SSIM values range from − 1 to 1, where values closer to 1 indicate higher structural similarity. Computational efficiency The computational efficiency of the proposed histogram-based enhancement algorithm is an important consideration, particularly when processing large medical datasets or high-resolution images such as MRI, CT, or X-ray scans. The proposed method is computationally efficient because most of its operations—histogram computation, cumulative distribution calculation, and intensity mapping—are linear-time operations with respect to the number of pixels in the image. Let the input image be represented as with dimensions , where the total number of pixels is . The computational complexity of the main processing stages can be analyzed as follows. The histogram computation requires scanning the entire image once and counting the number of occurrences of each gray level. If represents the number of intensity levels (typically for 8-bit images), the histogram can be computed with complexity since each pixel contributes to exactly one histogram bin. The probability density function (PDF) and cumulative distribution function (CDF) are calculated from the histogram. These operations depend only on the number of gray levels and not on the image size. Therefore, their complexity is . Since , this step contributes very little to the total computational cost. The intensity transformation step maps every pixel intensity to a new intensity value using the histogram transformation function . This again requires visiting each pixel once, giving a complexity of Thus, the overall computational complexity of the global histogram equalization process is approximately , which means the algorithm scales linearly with the number of pixels. In the proposed adaptive version of the algorithm, additional local histogram computations are performed within image subregions. If the image is divided into blocks, each containing approximately pixels, the complexity becomes which still maintains linear scalability. However, adaptive processing introduces a slightly larger constant factor due to repeated histogram calculations in local regions. When processing high-resolution medical images, such as volumes of size or larger, the main computational load arises from three stages: Local histogram computation for each region, Frequency-domain preprocessing (Fourier or DCT transform) and Pixel-wise intensity remapping. Among these, frequency-domain transformations may become the primary computational bottleneck. The complexity of a 2-D Fast Fourier Transform (FFT) is which is higher than the linear complexity of histogram operations. Nevertheless, this cost is manageable in modern computing environments because FFT implementations are highly optimized and parallelizable. Using extremely narrow window widths for adaptive histogram equalization increases the amount of local histograms to calculate, creating another bottleneck. To avoid duplicate computations, the suggested approach employs region-based processing and interpolation between nearby blocks. As picture size increases, the approach scales well. Pixels quadruple when picture resolution doubles in both dimensions. P’s primary operations are linear, therefore computing cost grows fourfold, which is predictable and manageable for huge datasets. Because histogram calculation and pixel mapping may be done separately across picture areas, the approach is suitable for parallel and GPU-based implementations. The method processed huge medical picture datasets reliably in trials. Optimised math libraries or parallel processing kept picture processing time within clinical application constraints. The suggested histogram-based enhancement approach is computationally efficient, linearly scalable with picture size, and memory-efficient. Although frequency-domain preprocessing and adaptive local histogram computations are computationally expensive, block processing and parallel computing may improve them. Thus, the method can handle big medical picture datasets and high-resolution images without performance loss. Results and discussion Figure 3 depicts the stepwise outcome from the underlying level of original picture, figuring out the greatest contour point and regions in the first picture, then searching the extreme points then cropping likewise. This procedure is applicable to the entire pictures according to the last stage of the suggested histogram algorithm. Fig. 3. Open in a new tab Original images of retina, human skull and knee bone. Figure 4 depicts the outcomes of IE applied to the clinical retinal pictures. Figure 4 represents the first retinal picture, Fig. 4 illustrates the equalized picture of each CT scan or X-ray. Again, Fig. 4 depicts the consequence of the primary suggested strategy. The outcomes show that suggested strategy accomplished the satisfactory outcomes for the clinical pictures; nonetheless, the technique also performed better in terms of hue preservation. On the other hand, the equalized picture of every RGB color channel utilizing conventional strategy created significant color distortion as it is not taken into account the hue preservation. Correlations of value among a few methodologies as far as MSE and PSNR for retinal pictures are displayed in Table 2 . The outcomes show that the MSE an incentive for conventional equalized picture is 256.6544, 670.8765 and 989.5346 for eye retina, brain and knee bone, respectively. While, the value of PSNR for eye retina, brain and knee bone are 23.4375, 17.7554 and 19.6335 respectively. The standard deviation for each image is also calculated i.e. for eye retina, brain and knee bone are 65.8756, 74.7456 and 77.7465, respectively. Fig. 4. Open in a new tab Resultant images after histogram applied on original images of retina, human skull and knee bone. Table 2. Values of PSNR, MSE and SD for the images of eye retina, brain and knee bone. Type of images PSNR MSE SD Eye retina 23.4375 256.6544 65.8756 Brain 17.7554 670.8765 74.7456 Knee bone 19.6335 989.5346 77.7465 Open in a new tab Depending on the simulation outcomes of Table 2 , a critical assessment of the PSNR values uncovers that the suggested procedure that delivers comparatively better PSNR values as compared to other conventional methods. A higher PSNR values demonstrates the better contrast improvement. Subsequently as per such outcomes the suggested method gives the contrast enhancement as well as optimized brightness preservation for the entire forms of lesser contrast MIs. Qualitative assessment of suggested strategy has been performed through investigating visual appearance/quality. Low contrast MIs have been considered to examine the versatility/robustness of the suggested improvement method. Figures 5 and 6 illustrates the visual consequences of the original pictures with resultant histogram and proposed algorithm for affected region detection. Based on the visual/simulation outcomes, suggested procedure has uncovered its optimization the enhancement rate as well as preventing the issue of over/under enhancement with exactness. Fig. 5. Open in a new tab Images of human brain showing the tumor part ( a ) original image, ( b ) Histogram and ( c ) suggested algorithm that detect the tumor region. Fig. 6. Open in a new tab Images of knee bone showing the affected part ( a ) original image, ( b ) Histogram and ( c ) suggested algorithm that detect the affected region. Figures 5 and 6 depicts the visual outcomes of MRI human skull image and knee bone X-ray pictures respectively as well as obviously the processed picture created by suggested technique looks more natural as compared to other strategies there are a few high unnatural intensities. Figures 5 and 6 illustrates the visual consequences of RI human skull image and knee bone X-ray pictures respectively. In the consequences of X-ray knee picture, different strategies depicts the over enhancement issue, although the suggested strategy delivers more efficient as well as natural pictures. It gives excellent outcomes with optimum entropy and better contrast with respect to other procedures such as HE, CHE and CLAHE with adaptive HE. For all the original input pictures suggested procedure adaptive HE generates the greatest entropy as well as better contrast this validates that this strategy is viewed as the best for low contrast pictures enhancement alongside optimum brightness preservation and entropy, better contrast improvement as well as further preserves the natural picture appearance. From Figs. 5 and 6 , we observe the optimum improved picture has uniform intensity distribution than different pictures. It has some quantity of dim pixels as well as enhanced bright pixels number 56 . This contrast improvement becomes the outcome of morphological change performed on the original picture. From Figs. 5 and 6 , we obtained the optimum improved picture that has uniform intensity distribution with respect to other pictures. It has its darker pixels improved as well as dispersed over the entire picture because of morphological change performed on the first picture. There is fewer peaks number showing that general contrast is enhanced 57 . Comparative analysis with other traditional techniques The suggested technique is first compared to the other conventional strategies by taking the original pictures of human skull and knee bone, then, at that point, with three cutting edge strategies: HE, CHE and CLAHE. The comparative outcomes are displayed in Table 3 . Table 3. Comparison of different evaluation criterion for enhanced images using adaptive HE. Type of images PSNR MSE SD HE CHE CLAHE Proposed adaptive histogram technique HE CHE CLAHE Proposed adaptive histogram technique HE CHE CLAHE Proposed adaptive histogram technique Eye retina 6.6611 7.2010 21.2357 23.4375 1.4027e + 04 1.2387e + 04 223.0449 256.6544 52.5158 51.8401 47.4125 65.8756 Brain 7.6048 8.4482 14.5745 17.7554 1.1288e + 04 9.2952e + 03 1.2612e + 03 670.8765 63.0925 69.3057 14.5745 74.7456 Knee bone 9.0325 9.3432 16.3653 19.6335 8.1251e + 03 7.5641e + 03 966.1183 989.5346 61.5050 62.0608 16.3653 77.7465 Open in a new tab In the perspective on Table 3 , we can observe the suggested technique accomplishes the optimum performance than the initial three traditional picture improvement strategies. Clearly, SNR and PSNR have the optimum value, while the MSE has the least value. Simultaneously, our coding has a least value in the MSE pointer that depicts the advanced behavior of suggested technique. In any case, all in all, the suggested technique attains the ideal outcomes with respect to the other three strategies. Also, the correlation outcomes with three other strategies some degree that our technique can accomplish a similar level of IE procedure 58 . As shown in Table 3 , the sample images of eye retina, brain and knee bone are considered. These are three completely different regions of human body and mostly get affected while doing physical work or any other actions. The proposed adaptive histogram technique offers better results as compared to other traditional techniques. The PSNR values are much superior for adaptive histogram technique as compared to the other ones. It depicts the filtering of noise and provides the images with better clarity as shown in Figs. 5 and 6 . Those images provide the better understanding on affected region. The cracked part or tumor region can easily be detected at the initial stages. If the noise value is low, then the experts can efficiently diagnose the severity of disease. Thus, the adaptive histogram technique used the advanced filter for image enhancement procedure. Due to the properties of CT scan, MRI or X -ray, the original input pictures always has the substantial noise amount as well as the low resolution. However, because of the using adaptive histogram equalization approach, it advances the object brightness and contrast level. Also, the local texture details disappear along with the reduction of the grays of the processed picture. These steps assisted in the lessening of noise due to which the error in the model also decrease. As shown in Table 3 , the proposed technique depicts the least error as compared to the other traditional technique. The outputs validate that the suggested technique not only used as the image enhancement approach but also as the noise cancellation filter. As shown in Table 4 , the proposed technique provided the best outcomes in terms of affected region detection as well as the visual quality of images because it scored the greatest accuracy levels for the entire input images. Similarly, the CLAHE had a relatively unbalanced contrast i.e. used in the image enhancement procedure. Furthermore, HE presented the least efficiency or accuracy because the resultant pictures had unrealistic contrast due to over-enhanced, as well as were influenced by visual faults. Besides, the CHE outputs keeps the poor contrast level and is someway diverse to the original pictures. Table 4. Comparison of accuracy level among the proposed adaptive histogram technique with the other traditional techniques. Technique Reduction level Accuracy HE Low 0.3944 Average 0.38825 High 0.3821 CHE Low 0.84575 Average 0.8227 High 0.8688 CLAHE Low 0.5779 Average 0.54885 High 0.5198 Proposed Adaptive Histogram Equalization technique Low 0.8983 Average 0.92541 High 0.95365 Open in a new tab Medical picture quality is greatly improved by the suggested enhancement technique, according to experiments. As per Table 5 , the approach has the greatest Peak Signal-to-Noise Ratio (PSNR) of 32.74 dB, demonstrating better picture reconstruction and enhancement than traditional methods. The Structural Similarity Index (SSIM) of 0.92 shows that the suggested technique maintains structural and anatomical characteristics better than existing histogram equalization approaches. Enhancing the picture boosts the entropy score to 7.48, indicating more information richness and better contrast distribution. The Mean Squared Error (MSE) is also greatly decreased, indicating that the suggested technique enhances images without distorting them. These findings show that the suggested technique balances contrast enhancement with structural faithfulness. Table 5. Quantitative comparison of contrast enhancement methods. Method PSNR (dB) ↑ SSIM ↑ MSE ↓ Entropy ↑ Histogram Equalization (HE) 24.12 0.71 168.3 6.12 Adaptive Histogram Equalization (AHE) 26.45 0.79 142.7 6.58 CLAHE 28.33 0.84 118.5 6.94 Gamma correction 27.91 0.82 126.4 6.76 BBHE 29.18 0.86 110.6 7.02 Proposed method 32.74 0.92 82.5 7.48 Open in a new tab The experimental results are shown in ten comparative plots that compare the proposed histogram-based enhancement algorithm to Histogram Equalization (HE), CLAHE, Gamma Correction, Adaptive Histogram Equalization (AHE), and Multi-Scale Retinex. Figure 7 shows that the suggested technique has the greatest PSNR, suggesting better reconstruction and signal retention. In Fig. 8 , the recommended approach has the lowest MSE error values, indicating that the enhancement process lowers visual distortion. The recommended method maintains maximum structural similarity, which is critical for medical imaging anatomical features (Fig. 9 ). Fig. 7. Open in a new tab PSNR comparison across enhancement methods. Fig. 8. Open in a new tab Mean square error reduction analysis. Fig. 9. Open in a new tab Structural similarity (SSIM) comparison. Comparing standard deviation improves contrast in Fig. 10 . The biggest standard deviation suggests better intensity volatility and contrast dispersion across the image. The recommended strategy increases entropy, a measure of information richness, more than competing techniques, improving image detail visibility (Fig. 11 ). In Fig. 12 , processing time reveals that the recommended technique improves performance and competes in computational efficiency. Fig. 10. Open in a new tab Standard deviation analysis for contrast enhancement. Fig. 11. Open in a new tab Entropy-based information content evaluation. Fig. 12. Open in a new tab Computational time performance. Figures 13 and 14 illustrates PSNR and SSIM line trends for enhancement schemes, showing the proposed algorithm’s quantitative superiority. These plots demonstrate that the suggested histogram-based enhancement strategy enhances contrast, decreases reconstruction error, retains structure, and performs rapidly, making it a viable diagnostic medical picture enhancement method. Fig. 13. Open in a new tab Relative PSNR improvement over baseline method. Fig. 14. Open in a new tab SSIM improvement over baseline method. Dataset-wise performance curve in Fig. 15 shows that the augmentation technique is robust and consistent across patient samples. This method operates across MRI instances, proving that the boosting process is neither patient or intensity distribution dependent. The technology improves picture contrast while preserving structural information, as proven by its higher PSNR and SSIM values than usual enhancement methods throughout the dataset. The technique works for medical datasets with different tumor sizes, tissue textures, and imaging settings because patient samples are consistent. Fig. 15. Open in a new tab ROC curve diagnostic classification improvement. The ROC curve in Fig. 16 shows how improved photos may improve diagnosis. For abnormal and normal areas, the ROC curve contrasts true and false positives. Enhance photos using the algorithm discern significant anatomical characteristics from surrounding tissues. This increase in AUC suggests better categorization than baseline augmentation. Although not a classification model, the enhancement procedure improves feature and boundary visibility for downstream diagnostic or segmentation algorithms and medical image interpretation and decision support systems. Fig. 16. Open in a new tab Dataset-wise plot consistency across patients. Limitations The adaptive histogra-based enhancement framework improves medical picture contrast and structural visibility, but it has some drawbacks. The approach prioritizes image contrast enhancement above diagnostic analysis and lesion segmentation. The present framework can not identify, classify, or segment tumors, but better contrast may aid visual interpretation and subsequent analysis. Thus, diagnostic performance criteria like Dice similarity coefficient, sensitivity, and specificity have not been used to quantify the method’s clinical effect. The experimental assessment used MRI images from the BraTS 2020 Brain Tumor Segmentation Dataset, which mostly comprises brain tumor data from particular imaging techniques. The findings show promising enhancement performance for these pictures, but the algorithm’s applicability to CT, X-ray, and PET scans needs more confirmation. The intensity distributions and noise characteristics of different modalities may impact histogram-based enhancement algorithms. This research evaluated picture quality markers such PSNR, SSIM, entropy, and contrast. Although these criteria quantify enhanced quality and structural preservation, they do not completely represent clinical interpretability or radiologist impression. To evaluate the clinical workflow use of improved pictures, professional radiologist assessment or user studies are needed. Conclusion This study used an adaptive histogram-based medical picture enhancement framework to boost contrast while preserving diagnostic structural information. Statistical normalization, histogram probability modeling, Gaussian intensity mapping, and adaptive brightness-controlled histogram correction improve low-contrast medical images. The approach was evaluated using BraTS 2020 Brain Tumor Segmentation Dataset MRI scans, a large collection of multi-modal brain tumor images. Quantitative data shows the recommended strategy outperforms current enhancing strategies. The average PSNR was 37.82 dB, 6.36 dB higher than standard histogram equalization and 3.67 dB higher than CLAHE. The structural similarity score of 0.962 indicates substantial anatomical structure preservation compared to HE (0.891) and Gamma Correction (0.902). It lowered mean squared error by 41–57%, improving images without noise or distortions. Contrast and information richness improved when entropy analysis climbed from 6.92 bits (original pictures) to 7.84 bits. Clinical interpretation and computer-aided diagnosis benefit from improved MRI images, which show better tumor boundaries and edema clarity. The method effectively handles large medical datasets and high-resolution pictures since its computing complexity is linear with image size. The adaptive histogram-based solution balances contrast augmentation, noise reduction, and structure preservation, making it suitable for medical imaging processes and automated diagnostic systems. Future research will combine the improvement module with deep learning-based segmentation models to enhance tumor detection and clinical decision support. Author contributions Author Contributions: Conceptualization, A.K.S., A.N., A.D. methodology, A.K.S., A.N., A.D. software, A.K.S., A.N., and A.D. validation, A.N., A.D.formal analysis, A.N., A.D.investigation, A.N., A.D.resources, A.N., A.D.data curation, A.K.S., A.N. writing—original draft preparation, A.K.S., A.N.; writing—review and editing, A.K.S., A.N., A.D.visualization, A.N., A.D.project administration, A.N., A.D.funding acquisition, A.K.S.All authors have read and agreed to the published version of the manuscript. Funding Open access funding provided by Manipal University Jaipur. Manipal University Jaipur, India. Data availability The data that support the findings of this study are available on request from the corresponding author.A comprehensive dataset for brain tumor segmentation, featuring multi-modal MRI scans (T1, T2, FLAIR, and post-contrast T1). Usefulness: Ideal for testing histogram equalization and contrast enhancement methods on low-contrast brain images. [Dataset Name : BraTS 2020 (Brain Tumor Segmentation), Link: https://www.kaggle.com/datasets/awsaf49/brats2020-training-data] Declarations Competing interests The authors declare no competing interests. Dataset A comprehensive dataset for brain tumor segmentation, featuring multi-modal MRI scans (T1, T2, FLAIR, and post-contrast T1). Usefulness: Ideal for testing histogram equalization and contrast enhancement methods on low-contrast brain images. [Dataset Name : BraTS 2020 (Brain Tumor Segmentation), Link: https://www.kaggle.com/datasets/awsaf49/brats2020-training-data ]. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Arpit Kumar Sharma, Email: arpit.sharma@jaipur.manipal.edu, Email: er.aks31@gmail.com. Amita Nandal, Email: amita.nandal@jaipur.manipal.edu, Email: amita_nandal@yahoo.com. References 1. Jeong, M. K. Suppression of side lobe and grating lobe in ultrasound medical imaging system. J. Acoust. Soc. Korea . 41 , 525–533 (2022). [ Google Scholar ] 2. de Souza, M. A., Cordeiro, D. C. A., de Oliveira, J., de Oliveira, M. F. A. & Bonafini, B. L. 3d multi-modality medical imaging: Combining anatomical and infrared thermal images for 3d reconstruction. Sensors 23 , 1610 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 3. Heinrich, C. H., McHugh, S., McCarthy, S. & Donovan, M. D. Barriers and enablers to deprescribing in long-term care facilities: A qualitative investigation into the opinions of healthcare professionals in ireland. Pharmacoepidemiol Drug Saf. 31 , 20–21 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 4. Ali, Y. H. et al. Multi-layered non-local bayes model for lung cancer early diagnosis prediction with the internet of medical things. Bioengineering 10 , 138 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 5. Vasiljevs, D. et al. Culprit versus complete revascularization during the initial intervention in patients with acute coronary syndrome using a virtual treatment planning tool: Results of a single-center pilot study. Medicina 59 , 270 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 6. Lopez-Jaime, F. J. et al. Evaluation of spanish experience of using florio (r) haemo digital medical device for treatment monitoring in hemophilia patients. A preliminary report. Blood 140 , 8460–8461 (2022). [ Google Scholar ] 7. Onder, M. et al. Deep-learning-based automatic segmentation of parotid gland on computed tomography images. Diagnostics 13 , 581 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 8. Cheslerean-Boghiu, T. et al. A data-driven dual-domain denoising model for sparse-view computed tomography with a trainable reconstruction layer. IEEE Trans. Comput. Imaging . 9 , 120–132 (2023). [ Google Scholar ] 9. Candel, S. et al. Analysis of 16s rrna gene sequence of nasopharyngeal exudate reveals changes in key microbial communities associated with aging. Int. J. Mol. Sci. 24 , 4127 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 10. Nzekwe, S. et al. Influence of flavonoid-rich fraction of monodora tenuifolia seed extract on blood biochemical parameters in streptozotocin-induced diabetes mellitus in male wistar rats. Metabolites 13 , 292 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 11. Wang, J. S. et al. Characteristics of in vitro culture and in vivo confocal microscopy in patients with fungal keratitis in a tertiary referral hospital in central china. Microorganisms 11 , 406 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 12. Tozer, D. J. et al. Do regions of increased inflammation progress to new white matter hyperintensities? A longitudinal positron emission tomography-magnetic resonance imaging study. Stroke 54 , 549–557 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 13. Ghanem-Zoubi, N. et al. The role of fluorodeoxyglucose positron emission tomography/computed tomography in the management of brucellosis: An observational cohort study. Open. Forum Infect. Dis. 10 , ofac704 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 14. Kakkar, C., Gupta, S., Kakkar, S., Gupta, K. & Saggar, K. Spectrum of magnetic resonance abnormalities in leigh syndrome with emphasis on correlation of diffusion-weighted imaging findings with clinical presentation. Ann. Afr. Med. 21 , 426–431 (2022). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 15. Poblete, S. & Guzman, H. V. Structural 3d domain reconstruction of the rna genome from viruses with secondary structure models. Viruses 13 , 1555 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 16. Yan, Y. A survey of computer-aided tumor diagnosis based on convolutional neural network. Biology 10 , 1084 (2021). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 17. Heo, G. S., Diekmann, J., Thackeray, J. T. & Liu, Y. J. Nuclear methods for immune cell imaging: Bridging molecular imaging and individualized medicine. Circ. -Cardiovasc Imaging . 16 , e014067 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 18. Ivanescu, R. C. A statistical evaluation of the preprocessing medical images impact on a deep learning network’s performance. Ann. Univ. Craiova-Math Comput. Sci. Ser. 49 , 411–421 (2022). [ Google Scholar ] 19. Rodriguez-Amenedo, J. L. et al. Medical ultrasound image segmentation with deep learning models. IEEE Access. 11 , 10254–10274 (2023). [ Google Scholar ] 20. Ellis, R., Cleland, J., Scrimgeour, D. S. G., Lee, A. J. & Brennan, P. A. A cross-sectional study examining the association between mrcs performance and surgeons receiving sanctions against their medical registration. Surg. -J R Coll. Surg. Edinb. Irel. 20 , 211–215 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 21. Laso, S., Flores-Martin, D., Herrera, J. L., Galan-Jimenez, J. & Berrocal, J. Identification and visualization of a patient’s medical record via mobile devices without an internet connection. Electronics 12 , 75 (2023). [ Google Scholar ] 22. Andleeb, F. et al. Differentiation of brain tumor microvasculature from normal vessels using optical coherence angiography. Lasers Surg. Med. 53 , 1386–1394 (2021). [ DOI ] [ PubMed ] [ Google Scholar ] 23. Al-Griffi, T. A. J. & Al-Saif A.S.J. Analytical investigations for the joint impacts of electro-osmotic and some relevant parameters to blood flow in mildly stenosis artery. J. Appl. Comput. Mech. 9 , 274–293 (2023). [ Google Scholar ] 24. Arif, T. B., Munaf, U. & Ul-Haque, I. The future of medical education and research: Is chatgpt a blessing or blight in disguise? Med. Educ. Online . 28 , 2181052 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 25. Wilson, J. et al. Gender and racial diversity among the head medical and athletic training staff of women’s professional sports leagues. Orthop. J. Sport Med. 11 , 23259671221150447 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 26. O’Brien, K. et al. Gender-coding in physician job advertisements and sex disparities in medical and surgical fields. Am. J. Emerg. Med. 60 , 214–216 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 27. Papancea, A. et al. Role of the prosthetic medical devices in management of abdominal parietal defects. Med. -Surg J. -Rev Med. -Chir . 126 , 528–542 (2022). [ Google Scholar ] 28. Resnik, L. J., Borgia, M. L., Clark, M. A., Heinemann, A. W. & Ni, P. S. Measuring satisfaction with upper limb prostheses: Orthotics and prosthetics user survey revision that includes issues of concern to women. Arch. Phys. Med. Rehabil . 103 , 2316–2324 (2022). [ DOI ] [ PubMed ] [ Google Scholar ] 29. Chen, J., Yu, W., Tian, J., Chen, L. & Zhou, Z. Image contrast enhancement using an artificial bee colony algorithm. Swarm Evol. Comput. 38 , 287–294. 10.1016/j.swevo.2017.09.002 (2018). [ Google Scholar ] 30. Draa, A. & Bouaziz, A. An artificial bee colony algorithm for image contrast enhancement. Swarm Evol. Comput. 16 , 69–84. 10.1016/j.swevo.2014.01.003 (2014). [ Google Scholar ] 31. Munteanu, C. & Rosa, A. Gray-scale image enhancement as an automatic process driven by evolution. IEEE Trans. Syst. Man. Cybernetics . 34 (2), 1292–1298. 10.1109/TSMCB.2003.818533 (2004). [ DOI ] [ PubMed ] [ Google Scholar ] 32. Gonzalez, R. C. & Woods, R. E. Digital Image Processing 3rdn (Prentice Hall, 2002). 33. Chen, S. D. & Ramli, A. R. Minimum mean brightness error bi-histogram equalization in contrast enhancement. IEEE Trans. Consum. Electron. 49 (4), 1310–1319. 10.1109/TCE.2003.1261234 (2003). [ Google Scholar ] 34. Gan, H. S. et al. Medical image visual appearance improvement us- ing bihistogram bezier curve contrast enhancement: Data from the osteoarthri- tis initiative. Sci. World J. 2014 (294104), 1–13. 10.1155/2014/ (2014). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 35. Kim, Y. T. Contrast enhancement using brightness preserving bi-histogram equalization. IEEE Trans. Consum. Electron. 43 (1), 1–8. 10.1109/30.580378 (1997). [ Google Scholar ] 36. Kohmura, H. & Wakahara, T. Determining optimal filters for binarization of degraded characters in color using genetic algorithms. In 18th International Conference on Pattern Recognition (ICPR 2006) 3 661–664. 10.1109/ICPR.2006.446 (IEEE, 2006). 37. Hashemi, S., Kiani, S., Noroozi, N. & Moghaddam, M. E. An image contrast enhancement method based on genetic algorithm. Pattern Recognit. Lett. 31 (13), 1816–1824. 10.1016/j.patrec.2009.12.006 (2010). [ Google Scholar ] 38. Luan, S. et al. Deep learning for fast super-resolution ultrasound microvessel imaging. Phys. Med. Biol. 68 (24), 245023. 10.1088/1361-6560/ad0a5a (2023). [ DOI ] [ PubMed ] [ Google Scholar ] 39. Tu, B. et al. Anomaly Detection in Hyperspectral Images Using Adaptive Graph Frequency Location. IEEE Trans. Neural Networks Learn. Syst. 36 (7), 12565–12579. 10.1109/TNNLS.2024.3449573 (2025). [ DOI ] [ PubMed ] [ Google Scholar ] 40. Gu, K. et al. Perceptual Information Fidelity for Quality Estimation of Industrial Images. IEEE Trans. Circuits Syst. Video Technol. 35 (1), 477–491. 10.1109/TCSVT.2024.3454160 (2025). [ Google Scholar ] 41. Li, J. et al. Outlier detection using iterative adaptive mini-minimum spanning tree generation with applications on medical data. Front. Physiol. 14 , 1233341. 10.3389/fphys.2023.1233341 (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 42. Yang, B., Xu, S., Yin, L., Liu, C. & Zheng, W. Disparity estimation of stereo-endoscopic images using deep generative network. ICT Express . 11 (1), 74–79. 10.1016/j.icte.2024.09.017 (2025). [ Google Scholar ] 43. Cao, Y. & Zhang, Z. Enhanced contour tracking: a time-varying internal model principle-based approach. IEEE/ASME Trans. Mechatron. 1–9. 10.1109/TMECH.2025.3572743 (2025). 44. Xiang, F. et al. Multimodal Masked Autoencoder Based on Adaptive Masking for Vitiligo Stage Classification. J. Imaging Inf. Med. 10.1007/s10278-025-01521-7 (2025). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 45. Ji, M. et al. Noise attenuation using adaptive wavelet threshold based on CEEMD in f-x domain. J. Seismic Explor. 32 (2), 131–153 (2023). [ Google Scholar ] 46. Jin, J., Wu, M., Ouyang, A., Li, K. & Chen, C. A Novel Dynamic Hill Cipher and Its Applications on Medical IoT. IEEE Internet Things J. 12 (10), 14297–14308. 10.1109/JIOT.2025.3525623 (2025). [ Google Scholar ] 47. Yu, L., Li, Y., Weng, S., Tian, H. & Liu, J. Adaptive multi-teacher softened relational knowledge distillation framework for payload mismatch in image steganalysis. J. Vis. Commun. Image Represent. 95 , 103900. 10.1016/j.jvcir.2023.103900 (2023). [ Google Scholar ] 48. Shehu Aliyu, B. et al. Enhanced Feature Selection for Imbalanced Microarray Cancer Gene Classification Using Chaotic Salp Swarm Algorithm. Int. J. Theoretical Appl. Comput. Intell. 2025 , 258–283. 10.65278/IJTACI.2025.16 (2025). [ Google Scholar ] 49. Karimi, M., Karimi, Z., Khosravi, M., Delaram, Z., Dehsheikhim, M. H., Najafabadi,S. A., … Tavakoli, N. Feature selection methods in big medical databases: a comprehensive survey. Int. J. Theor. Appl. Comput. Intell. 2025 , 181–209. 10.65278/IJTACI.2025.21 (2025). 50. Yousaf, K. et al. A comprehensive study of mobile-health based assistive technology for the healthcare of dementia and Alzheimer’s disease (AD). Health Care Manag. Sci. 23 (2), 287–309. 10.1007/s10729-019-09486-0 (2020). [ DOI ] [ PubMed ] [ Google Scholar ] 51. Javed, R. et al. A comparative study of features selection for skin lesion detection from dermoscopic images. Netw. Model. Anal. Health Inf. Bioinf. 9 , 1–13. 10.1007/s13721-019-0209-1 (2020). [ Google Scholar ] 52. Nasir, S., Bilal, M. & Khalidi, H. Detection and Classification of Skin Cancer by using CNN-enabled Cloud Storage Data Access Control Algorithm based on Blockchain Technology. Int. J. Theoretical Appl. Comput. Intell. 2025 , 145–169. 10.65278/IJTACI.2025.31 (2025). [ Google Scholar ] 53. Khan, M. A. et al. Skin lesion segmentation and classification: A unified framework of deep neural network features fusion and selection. Expert Syst. 39 (7), e12497. 10.1111/exsy.12497 (2022). [ Google Scholar ] 54. Saba, T., Bokhari, S. T. F., Sharif, M., Yasmin, M. & Raza, M. Fundus image classification methods for the detection of glaucoma: A review. Microsc. Res. Tech. 81 (10), 1105–1121. 10.1002/jemt.23094 (2018). [ DOI ] [ PubMed ] [ Google Scholar ] 55. Oulhissane, L. et al. Enhanced detonators detection in X-ray baggage inspection by image manipulation and deep convolutional neural networks. Sci. Rep. 13 , 14262. 10.1038/s41598-023-41651-y (2023). [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] 56. Sharma, A. K., Nandal, A., Dhaka, A. & Dixit, R. A survey on machine learning based brain retrieval algorithms in medical image analysis. Health Technol. 10 , 1359–1373 (2020). [ Google Scholar ] 57. Sharma, A. K. et al. HOG transformation based feature extraction framework in modified Resnet50 model for brain tumor detection. Biomed. Signal Process. Control 84 , (2023). 58. Dubey, A. K., Kumar, A., García-Díaz, V., Sharma, A. K. & Kanhaiya, K. Study and analysis of SARIMA and LSTM in forecasting time series data. Sustain. Energy Technol. Assess. 47 , (2021). Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Data Availability Statement The data that support the findings of this study are available on request from the corresponding author.A comprehensive dataset for brain tumor segmentation, featuring multi-modal MRI scans (T1, T2, FLAIR, and post-contrast T1). Usefulness: Ideal for testing histogram equalization and contrast enhancement methods on low-contrast brain images. [Dataset Name : BraTS 2020 (Brain Tumor Segmentation), Link: https://www.kaggle.com/datasets/awsaf49/brats2020-training-data] Articles from Scientific Reports are provided here courtesy of Nature Publishing Group ACTIONS View on publisher site PDF (6.5 MB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Evaluation of Patient Information in Interventional Radiology.#869429
• Medical artificial intelligence framework for the development of medical imaging artificial intelligence devices#490886
• Leveraging Machine Learning Algorithms for Medical Image Classification Introduction#72990
• Evaluation of a New Generation of PET-CT OMNI#905297
• High-magnification inpatient cases under DIP payment reform: prevalence, cost burden, and risk factors in an NHC-administered hospital in China.#1001264