Interpretable Multiple-Instance Learning of Glomerular Pathomics and Subtype Ratios for eGFR Classification
Authors: Tianyu Shi, Yuechen Yang, Leiyue Zhao, Daniel Reisenbüchler, Jing Jin, Thangamani Muthukumar, Kenji Ikemura, Mert R. Sabuncu, Yihe Yang, Ruining Deng
Organizations: Sichuan University, Chengdu, 610065, CN · Vanderbilt University, Nashville, TN, 37235, USA · Johns Hopkins University, Baltimore, MD 21218, USA · University of Regensburg, Regensburg, Bavaria 93053, DE · Northwestern University, Evanston, IL 60208, USA · Weill Medical College of Cornell University, New York, NY 10065, USA · Cornell Tech, New York, NY 10044, USA · Northwell Health, New Hyde Park, NY 11040, USA
Chronic kidney disease is characterized by structural alterations in renal tissue and progressive loss of kidney function. While glomerular subtype ratios provide a compact and clinically meaningful summary of glomerular composition, representing pathomic information using whole-slide image (WSI)-level mean values diminishes glomerulus-level heterogeneity and the contribution of less prevalent patterns. In this proof-of-concept study, we investigated whether modeling individual glomeruli using multiple-instance learning (MIL) could provide complementary information to conventional subtype ratios for estimated glomerular filtration rate (eGFR) classification. We analyzed 162 kidney-biopsy WSIs from 162 patients, categorizing renal function as eGFR < 60 or >= 60 mL/min/1.73 m^2. Quantitative size, shape, intensity, and texture features were extracted from segmented glomeruli, representing each patient as a bag of glomerular instances. An attention-based MIL model was trained to aggregate these heterogeneous pathomic profiles. We evaluated subtype ratios, pathomics, and their combination using ABMIL, with a mean-pooled multilayer perceptron baseline. In five-fold cross-validation, the subtype-ratio-only model achieved an AUC of 0.7262 +/- 0.0596. MIL using pathomics alone achieved an AUC of 0.7362 +/- 0.0788, whereas integrating pathomics with subtype ratios produced the highest mean AUC of 0.7630 +/- 0.0665. SHapley Additive exPlanations (SHAP) were used to examine individual feature contributions. These results demonstrate that MIL-based pathomics can complement established glomerular subtype ratios by capturing additional glomerulus-level information for quantitative kidney-biopsy analysis.
Figures & tables
Figure 1: Schematic overview of the three experimental settings evaluated for eGFR classification. (A) Subtype Ratios: Uses only the global proportions of glomerular subtypes (e.g., viable, ischemic, globally sclerotic) as the patient-level feature. (B) Mean Pooling: A baseline architecture that extracts pathomics features from individual glomeruli and aggregates them via uniform mean pooling to generate a patient-level representation, treating all instances equally. (C) Attention-based MIL: The proposed pipeline, which performs random instance sampling ( ns=256 ), instance encoding, and gated attention pooling to generate a weighted pathomics representation. This representation is late-fused with the glomerular subtype ratios to yield the final eGFR prediction. In panel (C), the fusion symbol represents the learnable logit-level combination defined in Eq. 2 .
Figure 2: Overview of the proposed pipeline for predicting eGFR categories. The framework comprises two parallel branches. The Pathomics branch first performs random instance sampling ( ns=256 ) for each patient, followed by an instance encoder and gated attention pooling to aggregate instance-level embeddings into a patient-level representation. This representation is passed through a dedicated classifier to produce Pathomics Logits. The Glomerular subtype ratios branch processes the global subtype ratio features through a multi-layer perceptron (Linear, ReLU, Dropout) to generate Ratio Logits, which are then combined with the Pathomics Logits via a learnable convex combination. The fused logits are then passed through a softmax function to yield the final eGFR classification ( ≥60 vs. <60 ). The SHAP block denotes a separate post hoc attribution analysis of the fixed-dimensional attention-aggregated representation. It is not part of the prediction pathway and should not be interpreted as an exact decomposition of the complete ABMIL bag-level decision function.
Figure 3: Representative examples and group-level feature attribution for the observed eGFR≥60\penaltymL/min/1.73m2 ( n=88 ) and eGFR<60\penaltymL/min/1.73m2 ( n=74 ) categories. For each category, the left panel shows a representative WSI, the middle panel shows the nine glomeruli receiving the highest attention weights, and the right panels show the ten highest-ranked features. Bar lengths represent mean absolute SHAP values, and error bars indicate the standard error of the mean (SEM). SHAP summary plots show the distribution and direction of feature contributions across patients, with red and blue indicating relatively high and low feature values, respectively.
Figure 4: Group-level feature attribution analysis stratified by the percentage of globally sclerotic glomeruli (GGS). Three subgroups are evaluated: GGS <25% , 25%≤ GGS <50% , and GGS ≥50% . For each subgroup, the top 10 features are displayed. Left (Bar plots): Bars represent the mean absolute SHAP values, and error bars indicate the standard error of the mean (SEM), calculated as the standard deviation divided by the square root of the sample size ( n ). Right (Scatter plots): SHAP summary (beeswarm) plots show the distribution and direction of feature contributions across individual patients, where point color indicates the relative feature value (red: high, blue: low). To facilitate direct visual comparison across the GGS strata, the x-axis scales within each plot type (bar and scatter) are held constant across all three subgroups.
Glomerular pathology is central to the diagnosis and prognosis of renal diseases, yet the heterogeneity of glomerular morphology and fine-grained lesion patterns remain challenging for current AI approaches. We present GloPath, an entity-centric foundation model trained on over one million glomeruli extracted from 14,049 renal biopsy specimens using multi-scale and multi-view self-supervised learning. GloPath addresses two major challenges in nephropathology: glomerular lesion assessment and clinicopathological insights discovery. For lesion assessment, GloPath was benchmarked across three independent cohorts on 52 tasks, including lesion recognition, grading, few-shot classification, and cross-modality diagnosis-outperforming state-of-the-art methods in 42 tasks (80.8%). In the large-scale real-world study, it achieved an ROC-AUC of 91.51% for lesion recognition, demonstrating strong robustness in routine clinical settings. For clinicopathological insights, GloPath systematically revealed statistically significant associations between glomerular morphological parameters and clinical indicators across 224 morphology-clinical variable pairs, demonstrating its capacity to connect tissue-level pathology with patient-level outcomes. Together, these results position GloPath as a scalable and interpretable platform for glomerular lesion assessment and clinicopathological discovery, representing a step toward clinically translatable AI in renal pathology.
Qiming He, Jing Li, Tian Guan +26
Institute of Biopharmaceutical and Health Engineering, Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China · Interdisciplinary Institute for Medical Engineering, Fuzhou University, Fuzhou, China · Medical Optical Technology R&D Center, Research Institute of Tsinghua, Pearl River Delta, Guangzhou, China +6
Fine-grained evaluation of glomerular pathology must distinguish normal glomeruli from abnormalities such as global and segmental glomerulosclerosis, obsolescent, ischemic, solidified, disappearing, and atubular glomeruli. Supervised classification requires labeled examples of every category, which is impractical when subtypes are rare or absent from the training cohort. One-class anomaly detection offers an alternative by modeling normal data and scoring deviations, allowing previously unseen abnormalities to be detected. We use the frozen residual U-Net backbone of Omni-Seg, pretrained to segment structurally normal renal primitives without abnormal-subtype labels. We propose NoRDeC (Normal-Reference Detection and Characterization), a framework combining Mahalanobis normal-reference scoring with layer-wise representation analysis to determine whether and where glomerular pathology is encoded, how spatial aggregation affects detection, and whether abnormalities alter inter-layer relationships differently. Using glomerular images from two institutions, we evaluate backbone layers and aggregation strategies, compare NoRDeC with PaDiM and PatchCore, and analyze representations using centered kernel alignment (CKA). Layer 4 with Center-70 aggregation achieved a pooled AUROC of 0.926±0.013. NoRDeC achieved the highest AUROC in six of seven abnormality categories and in the pooled analysis, while CKA suggested subtype-dependent changes in inter-layer relationships not captured by anomaly scores alone. The normal-reference model is fitted using only normal glomeruli; abnormality labels are used for configuration selection, evaluation, and grouping in the representation analysis. These results show that a frozen renal feature extractor can support both detection and representation-level characterization of glomerular abnormalities without using abnormal examples to fit the detector.
Greta Hasko, Rachit Saluja, Tianyu Shi +11
Cornell University, Ithaca, NY 14853, USA · Cornell Tech, New York, NY 10044, USA · Sichuan University, Chengdu, 610207, CN +6
Fine-grained glomerular subtyping is essential for kidney biopsy interpretation, yet it presents a harder setting than the broad organ- or disease-level recognition tasks commonly used in vision-language model (VLM) evaluation. In this task, all categories arise within the same renal microanatomic unit, and clinically meaningful subtypes often differ by subtle, overlapping, and continuum-like morphological cues. Clinically valuable labels are also scarce, making fully supervised image-only paradigms difficult to translate to practice. Although VLMs are promising for data-efficient pathology learning, it remains unclear how their cross-modal representations should be adapted and evaluated for such within-structure discrimination under realistic few-shot constraints. To our knowledge, this study establishes the first fine-grained renal pathology benchmark for few-shot VLM adaptation and systematically examines how supervision level, model design, and adaptation strategy shape both classification performance and multimodal representation structure. Beyond accuracy, AUC, and F1, we assess image-text alignment, positive-negative similarity separation, embedding geometry, and calibration (ECE), asking when performance gains reflect more reliable cross-modal understanding. Across renal pathology cohorts, additional supervision consistently improves discrimination, but reliable adaptation is not explained by alignment alone; stronger performance is associated with larger class separation, more coherent embedding organization, and better awareness of confidence-related failure modes under domain variation. Within this retrospective two-cohort evaluation, these findings move assessment beyond coarse classification and accuracy alone and provide comparative evidence to inform model selection and future translational validation. Code: https://github.com/ddrrnn123/glo-vlms
Zhenhao Guo, Rachit Saluja, Tianyu Shi +13
New York University, New York, NY 10012, USA · Cornell Tech, New York, NY 10044, USA · Vanderbilt University, Nashville, TN, 37235, USA +4