An AI-based standardized automatic reporting system for breast cancer using HE whole slide images

HUANG Jie; MA Yong; RUAN Miao; FEI Xiaochun; ZHU Lifeng; DA Qian

doi:10.3969/j.issn.1007-8096.2025.12.007

Abstract：Objective To explore the application value of vision foundation models in intelligent diagnosis of whole slide images(WSI)for breast cancer,to construct an artificial intelligence-driven pathological visualization diagnostic platform,and to develop a standardized report auto-generation system,thereby improving the accuracy,standardization,and efficiency of breast cancer pathological diagnosis.Methods A total of 2,938 WSI images of breast resection specimens and corresponding standardized pathological reports were retrospectively collected from the Department of Pathology,Ruijin Hospital,Shanghai Jiaotong University School of Medicine,between January 2023 and 2025.The dataset was divided into training,validation,and test sets at a ratio of 7:1:2.The RuiPath Pathology Vision Foundation Model v1.0,independently developed by Ruijin Hospital and based on the Vision Transformer(ViT)architecture,was employed as the visual encoder.During training,regions of interest within the WSI were extracted and segmented into 256×256 pixel image patches for model input.During testing,the model's performance was evaluated on key diagnostic tasks,including tumor detection(presence or absence),histological typing,Nottingham grading(mitotic count,tubule formation,and nuclear pleomorphism),and lymph node metastasis.The primary evaluation metrics included F1 score,accuracy(ACC),and area under the curve(AUC).Finally,a web-based visual interactive system with a review function was developed to achieve full automation from image analysis to report generation.Results The pathological visualization diagnostic system demonstrated excellent performance in WSI-level tumor detection,accurately identifying the presence or absence of tumors within breast anatomical structures(nipple,base and quadrants).It presented probability heatmaps of tumor regions and visual annotations of key pathological features.The generated standardized reports included structured diagnostic elements,such as tumor size,histological type,grade,margin status,and lymph node metastasis,supporting multidimensional interactive review.Conclusion The artificial intelligence-assisted diagnostic system based on the ViT architecture vision foundation model can efficiently integrate multi-scale features from WSI,enabling automation and standardization of breast cancer pathological diagnosis and significantly improving diagnostic efficiency.The study confirms the clinical application potential of this technology in assisting pathologists with breast cancer pathological evaluation and report generation,offering a new paradigm for the advancement of digital pathology.Future multi-center prospective studies are required to further validate its clinical application efficacy.