Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types

2025-10-13

AI Summary

This study addresses the limited cross-cancer generalizability of tumor segmentation in whole-slide images (WSIs). We propose the first universal single-model solution, built upon a deep learning semantic segmentation framework and trained on over 20,000 multi-center, multi-scanner, multi-preparation WSIs spanning colorectal, endometrial, lung, and prostate cancers—without cancer-specific fine-tuning. The model achieves Dice scores >80% across six external validation cancer types and the TCGA dataset, matching the performance of cancer-specific models. Our key contribution is the first empirical demonstration that a single model can maintain high robustness and generalizability across clinically significant heterogeneity—including diverse cancer types, patient populations, scanner models, and staining protocols. This establishes a reproducible technical paradigm and benchmark for universal AI in digital pathology.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

Deep learning is expected to aid pathologists by automating tasks such as tumour segmentation. We aimed to develop one universal tumour segmentation model for histopathological images and examine its performance in different cancer types. The model was developed using over 20 000 whole-slide images from over 4 000 patients with colorectal, endometrial, lung, or prostate carcinoma. Performance was validated in pre-planned analyses on external cohorts with over 3 000 patients across six cancer types. Exploratory analyses included over 1 500 additional patients from The Cancer Genome Atlas. Average Dice coefficient was over 80% in all validation cohorts with en bloc resection specimens and in The Cancer Genome Atlas cohorts. No loss of performance was observed when comparing the universal model with models specialised on single cancer types. In conclusion, extensive and rigorous evaluations demonstrate that generic tumour segmentation by a single model is possible across cancer types, patient populations, sample preparations, and slide scanners.

Problem

• Develop universal tumor segmentation model for histopathology
• Validate model performance across multiple cancer types
• Compare universal model with cancer-specific segmentation models

Innovation

• Universal deep learning model for multi-cancer segmentation
• Trained on 20000+ whole-slide images across four cancers
• Validated on 3000+ external patients across six cancer types