Frontiers in medicine · 2026
Explainable AI for classifying vertebral fracture histology in digital spine pathology.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Yang Panyi
Аннотация
INTRODUCTION: Vertebral compression fractures (VCFs) commonly arise from osteoporosis, trauma, or malignancy. Accurate subtype differentiation is clinically essential but remains challenging using conventional imaging and histology.
METHODS: We developed an explainable AI-driven digital pathology pipeline integrating whole-slide histopathology with clinical metadata and transcriptomic profiles to support fracture subtype classification, risk stratification, and therapy prediction. Model interpretability was assessed using Grad-CAM heatmaps and SHAP analysis, and the multi-omics risk score was validated across independent cohorts.
RESULTS: The deep learning classifier achieved 86-91% accuracy (F1 score 0.83-0.88) for osteoporotic, traumatic, and neoplastic fractures, despite modest per-class AUCs (0.49-0.54). Grad-CAM and SHAP highlighted biologically meaningful cues, including trabecular thinning, nuclear atypia, and marrow fibrosis. The multi-omics risk score stratified outcomes: high-risk fractures showed upregulated TNF-NF-κB signaling, reduced cytotoxic T-cell infiltration, and significantly worse 3-year survival (log-rank p < 0.001). Drug sensitivity modeling predicted response patterns, with low-risk patients aligned with bisphosphonates and RANKL inhibitors and high-risk cases associated with resistant phenotypes.
DISCUSSION: This pipeline combines diagnostic performance with transparent interpretability and operates efficiently on modest computing resources, supporting telepathology deployment in resource-limited settings. By uniting classification, biological insight, and scalable implementation, the framework advances AI-enabled digital pathology toward more equitable global healthcare delivery.
METHODS: We developed an explainable AI-driven digital pathology pipeline integrating whole-slide histopathology with clinical metadata and transcriptomic profiles to support fracture subtype classification, risk stratification, and therapy prediction. Model interpretability was assessed using Grad-CAM heatmaps and SHAP analysis, and the multi-omics risk score was validated across independent cohorts.
RESULTS: The deep learning classifier achieved 86-91% accuracy (F1 score 0.83-0.88) for osteoporotic, traumatic, and neoplastic fractures, despite modest per-class AUCs (0.49-0.54). Grad-CAM and SHAP highlighted biologically meaningful cues, including trabecular thinning, nuclear atypia, and marrow fibrosis. The multi-omics risk score stratified outcomes: high-risk fractures showed upregulated TNF-NF-κB signaling, reduced cytotoxic T-cell infiltration, and significantly worse 3-year survival (log-rank p < 0.001). Drug sensitivity modeling predicted response patterns, with low-risk patients aligned with bisphosphonates and RANKL inhibitors and high-risk cases associated with resistant phenotypes.
DISCUSSION: This pipeline combines diagnostic performance with transparent interpretability and operates efficiently on modest computing resources, supporting telepathology deployment in resource-limited settings. By uniting classification, biological insight, and scalable implementation, the framework advances AI-enabled digital pathology toward more equitable global healthcare delivery.