The research, led by B. V. Divyashree of the Mysore University School of Engineering together with Shamair Khushba and G. Hemantha Kumar of the University of Mysore, deliberately charts a middle course between two dominant paradigms in digital pathology. On one side stand handcrafted, interpretable feature-based methods; on the other, data-hungry deep learning architectures such as convolutional neural networks, ResNet, DenseNet, and EfficientNet. Deep models have produced impressive results, but they typically demand large annotated datasets, substantial GPU memory, and careful hyperparameter tuning. By contrast, the pipeline proposed by the Mysore team relies on classical image processing and machine learning, offering transparency and computational efficiency that make it attractive for laboratories and clinics where data or hardware are limited.

The pipeline unfolds in three hierarchical stages: preprocessing, nuclei segmentation, and classification. Preprocessing itself begins with enhancement. The authors tested two complementary techniques on the raw images. The first is Gaussian filtering, a noise-reduction method that convolves the image with a kernel whose weights follow a Gaussian distribution, assigning the greatest weight to pixels near the kernel’s center. Because histopathological micrographs frequently contain Gaussian noise, this filter suppresses high-frequency artifacts while preserving the global intensity variations needed for later segmentation. The second technique, decorrelation stretch, addresses a different problem: in most histopathology images the red, green, and blue color channels are highly correlated and carry overlapping information, muting the color contrast that carries diagnostic meaning. The method rotates and rescales the pixel data into a new space where the channels become less correlated, then applies an inverse transformation back to the original space, producing brighter, more spectrally distinct images.

To decide objectively which enhancement method deserved a place in the pipeline, the researchers evaluated both using two quantitative metrics. Peak Signal-to-Noise Ratio, or PSNR, expresses the relationship between the maximum power of a signal and the power of the corrupting noise, while Root Mean Square Error, or RMSE, quantifies the pixel-wise difference between the enhanced image and the original reference. Higher PSNR and lower RMSE indicate better quality. Across images captured at magnifications of 40X, 100X, 200X, and 400X, and for both benign and malignant samples, the Gaussian filter delivered the highest PSNR and the lowest RMSE. Although decorrelation stretch produced visually stronger color separation, it could not match the Gaussian filter’s fidelity, so Gaussian-filtered images were carried forward into the segmentation stage.

The next challenge was isolating cell nuclei, the structures whose shape, size, and texture betray whether tissue is healthy or cancerous. In normal tissue, nuclei tend to be round and uniform; in malignant tissue they appear irregular and vary widely in size. Poor segmentation that bleeds in surrounding cytoplasm or stroma, or that merges overlapping nuclei, can distort these cues and doom the classifier downstream. The authors’ approach begins by converting the enhanced RGB image into the YCbCr color space, which decouples luminance, the Y channel representing brightness, from chrominance, the Cb and Cr channels encoding blue-difference and red-difference respectively. This separation makes segmentation robust to the staining variations and lighting inconsistencies that plague routine slides. Contrast Limited Adaptive Histogram Equalization, or CLAHE, is then applied only to the Y component. Unlike conventional histogram equalization, CLAHE operates on small local tiles and applies a contrast-limiting function, sharpening local detail without amplifying noise or over-enhancing the image, while the untouched chrominance channels preserve color fidelity.

With contrast enhanced, the pipeline applies K-means clustering to partition each pixel into one of three clusters, a value chosen empirically. The algorithm iteratively assigns pixels to the nearest cluster centroid and recomputes centroids until convergence, effectively dividing the image into background, normal tissue, and suspicious regions. One cluster corresponds to the nuclei, which are thereby separated from stroma, cytoplasm, and other cellular components. The authors report that the segmentation shows consistent uniformity and strong correlation with the underlying tissue structures, though regions where nuclei overlap remain a recognized challenge. Notably, because the BreakHis dataset used for validation provides ground-truth labels only for classification and not for segmentation, the segmentation module was assessed qualitatively rather than against annotated masks.

From the segmented nuclei, the team extracted a rich set of handcrafted features spanning three families. Texture features were computed from the Gray-Level Co-Occurrence Matrix, or GLCM, using two matrices that capture horizontal and vertical pixel relationships; these yield quantitative descriptors such as homogeneity, energy, correlation, contrast, and entropy that summarize the spatial arrangement of cells within tissue structures. Geometric features, including area, diameter, radius, perimeter, eccentricity, elongation, and solidity, characterize the shape variability that distinguishes cancerous nuclei from their benign counterparts. Finally, color features were drawn from the Cb and Cr components of the enhanced YCbCr images. The authors emphasize that these interpretable, handcrafted features were chosen deliberately over learned representations because they explicitly characterize tissue properties, enhance clinical explainability, and remain effective in limited-data scenarios where deep networks would struggle.

Classification was performed with two of machine learning’s most dependable workhorses: Support Vector Machine and K-Nearest Neighbor. The SVM, equipped with a radial basis function kernel selected empirically for its ability to capture non-linear relationships in the feature space, achieved the stronger results, reaching approximately 87.5 percent accuracy on the BreakHis dataset and outperforming both the K-NN baseline and many prior methods in the literature. For K-NN, the number of neighbors was varied from 1 to 15, with optimal performance at k equal to 5 using Euclidean distance, a setting that balances the low bias but high variance of small k against the increased bias of large k. Critically, the researchers adopted a patient-level split strategy: because BreakHis contains multiple images at different magnifications from the same patient, all images from a single patient were assigned exclusively to either the training or testing set, in an 80:20 ratio, preventing data leakage that can inflate reported accuracy. Performance was assessed with accuracy, sensitivity, specificity, precision, recall, and F-score.

The implications reach beyond a single benchmark score. In head-to-head comparison with representative prior work on BreakHis, the proposed pipeline achieved the highest reported accuracy while offering improved interpretability, lower computational complexity, and suitability for small-scale datasets without extensive GPU resources. An ablation study varying filtering techniques, color spaces, and clustering parameters confirmed that the integration of Gaussian enhancement, YCbCr conversion with CLAHE, and three-cluster K-means segmentation justifies each design choice. The authors acknowledge the approach’s limits: handcrafted features may miss highly complex patterns, and some nuclei remain over-segmented, leaving room for error reduction. Future work will integrate deep learning-based feature extraction for end-to-end learning and test robustness on additional datasets. For now, the study offers a compelling demonstration that careful, interpretable engineering, rather than sheer model size, can push computer-aided diagnosis of breast cancer meaningfully forward, with the potential to speed up pathology workflows and reduce diagnostic variability where it matters most.

Subject of Research: Machine learning classification of breast histopathology images into benign and malignant using a hierarchical preprocessing and segmentation pipeline

Article Title: A hierarchical pipeline for preprocessing and classification of breast histopathology images into benign and malignant

Article References: Divyashree, B. V., Khushba, S., & Kumar, G. H. (2026). A hierarchical pipeline for preprocessing and classification of breast histopathology images into benign and malignant. Discover Informatics, 1(1), Article 11. https://doi.org/10.1007/s44564-026-00013-2

Image Credits: AI Generated

DOI: 10.1007/s44564-026-00013-2

Keywords: breast cancer, histopathology, machine learning, SVM, K-means clustering, image segmentation, CLAHE, YCbCr, Gaussian filter, BreakHis dataset, computer-aided diagnosis, medical imaging

A team of researchers at Guangdong University of Technology, led by Bo Liu of the School of Automation, has now tackled this practical weakness head-on. In a study published in Neural Computing and Applications, the authors introduce a framework called Explainability-Guided Partial Privileged Learning, abbreviated EPPL, which is designed to keep LUPI methods effective even when only a fraction of the privileged information is actually available. The work addresses a scenario that is far more common in industry and medicine than the idealized complete-privileged-information setting: some training samples come with the extra knowledge, and many do not, and nobody has the resources to fill in all the gaps.

The core challenge the researchers identified is how to reconstruct the missing privileged information without contaminating the learning process with noise. A natural strategy is similarity transfer: if two training examples look alike in their ordinary feature space, one might infer that their privileged descriptions are also similar, allowing the privileged representation of a fully documented example to stand in for a sparsely documented one. The trouble is that naive similarity calculations in high-dimensional spaces are notoriously fragile. Two samples may appear close together because of background texture, lighting artifacts, or other irrelevant signals, while being semantically far apart in terms of what actually distinguishes their classes. This semantic misalignment means that blind similarity transfer can inject misleading surrogate information into the training pipeline, degrading rather than improving the final classifier.

EPPL’s central innovation is to stop being blind. The framework integrates an Explainable AI mechanism, specifically LIME, the Local Interpretable Model-agnostic Explanations technique introduced by Marco Ribeiro and colleagues in 2016. LIME probes a classifier by perturbing its inputs and observing how predictions change, producing a local explanation that highlights which features actually drove a given decision. In EPPL, these LIME-derived masks act as filters that separate discriminative features, the ones that genuinely matter for classification, from background noise. When the system computes similarity between two training samples to transfer privileged information, it weights the comparison by these masks, so that the distance is dominated by features the classifier itself has identified as meaningful rather than by incidental similarities across the full high-dimensional representation.

On top of this explanation-guided similarity measure, the researchers designed a class-aware weighted similarity strategy with two jobs. First, it improves the accuracy of imputing missing privileged information, because samples of the same class tend to share relevant privileged structure that class-conscious matching can exploit. Second, and crucially, the same similarity machinery dynamically assigns confidence weights to the downstream SVM+ classifier. SVM+, Vapnik’s extension of the support vector machine to the LUPI setting, learns a correcting function over the privileged space alongside the main decision function. If the imputed privileged information for a given sample is unreliable, treating it with the same confidence as genuinely observed data would be a mistake. EPPL instead down-weights samples whose reconstructed privileged information rests on weak similarity evidence, letting the classifier lean harder on well-supported examples and more gently on uncertain ones.

The quantitative results reported in the study are striking, particularly in the hardest regime. When only 20 percent of the privileged information was available, a severe data-scarce scenario, EPPL outperformed state-of-the-art baseline methods by up to 3.12 percent in classification accuracy. Equally important is the framework’s stability. As the proportion of available privileged information dropped from 80 percent to 20 percent, EPPL’s performance degraded by only 9.32 percent, while other competitive models suffered drops exceeding 12 percent. In a field where the gap between a good model and a deployable model is often measured in fractions of a percentage point, both the accuracy advantage and the reduced degradation represent meaningful progress.

The experiments were conducted on four publicly available benchmark datasets spanning very different domains: the Handwritten Digital Images dataset from the UCI Machine Learning Repository, the Caltech-101 object recognition collection, the Animals with Attributes 2 dataset, and the Corel image dataset. Testing across handwritten digits, natural object photographs, animal categories, and generic imagery gives the results a breadth that single-dataset studies often lack, and the consistent advantage of EPPL across these benchmarks suggests that the benefit derives from the framework itself rather than from idiosyncrasies of any one data domain. All four datasets are freely accessible, which should make it straightforward for other groups to reproduce and extend the findings.

Beyond the headline numbers, the study carries a broader conceptual message about the relationship between explainability and learning efficiency. Explainable AI is usually framed as a tool for human trust: producing heatmaps, feature rankings, and textual justifications so that people can audit what a model is doing. EPPL repurposes those same explanations for a purely technical role, using them as a signal that improves the learning pipeline itself. The LIME masks do not exist to reassure a clinician or a regulator; they exist to decide which pixel or feature dimensions should carry weight when the algorithm judges whether two training examples are true semantic neighbors. This inversion, in which explanations become functional components rather than post-hoc reporting layers, hints at a productive research direction where interpretability methods pay for themselves in raw performance.

The practical implications reach into any domain where privileged information is expensive. Medical imaging is an obvious candidate: expert annotations, additional imaging modalities, and longitudinal follow-up data all function as privileged information that cannot be gathered for every case. Financial modeling is another, where analysts’ private assessments or delayed regulatory filings provide training-time knowledge that is unavailable in real time. In settings such as these, the robustness curve matters as much as the peak. A method that performs brilliantly with complete information but craters when coverage drops to a fifth of the dataset forces practitioners into painful trade-offs between data budgets and model quality. EPPL’s comparatively gentle degradation suggests that partial coverage can be an acceptable operating point rather than a failure mode.

The research was supported by the Natural Science Foundation of China under Grant 62076074 and the Natural Science Foundation of Guangdong Province under Grant 2023A1515012560. The source code is available from the corresponding author upon reasonable request, and the authors report no competing financial interests. As privileged learning matures from a theoretical curiosity into a practical toolkit, the question of incomplete information is the question that will decide its real-world reach, and this work offers a concrete, explanation-driven answer to it.

Subject of Research: Explainability-guided partial privileged learning using class-aware weighted similarity to handle incomplete privileged information in machine learning classifiers

Article Title: Explainability-guided partial privileged learning: a class-aware weighted similarity approach

Article References: Explainability-guided partial privileged learning: a class-aware weighted similarity approach. (n.d.). https://doi.org/10.1007/s00521-026-12316-4

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12316-4

Keywords: privileged information, LUPI, explainable AI, LIME, SVM+, similarity transfer, machine learning, classification, partial privileged learning, Neural Computing and Applications, Guangdong University of Technology, data imputation