AI-augmented pathology: the experience of transfer learning and intra-domain data diversity in breast cancer metastasis detection

Abstract

Background:

Metastatic detection in sentinel lymph nodes remains a crucial prognostic factor in breast cancer management, with accurate and timely diagnosis directly impacting treatment decisions. While traditional histopathological assessment relies on microscopic examination of stained tissues, the digitization of slides as whole-slide images (WSI) has enabled the development of computer-aided diagnostic systems. These automated approaches offer potential improvements in detection consistency and efficiency compared to conventional methods.

Results:

This study leverages transfer learning on hematoxylin and eosin (HE) WSIs to achieve computationally efficient metastasis detection without compromising accuracy. We propose an approach for generating segmentation masks by transferring spatial annotations from immunohistochemistry (IHC) WSIs to corresponding H&E slides. Using these masks, four distinct datasets were constructed to fine-tune a pretrained ResNet50 model across eight different configurations, incorporating varied dataset combinations and data augmentation techniques. To enhance interpretability, we developed a visualization tool that employs color-coded probability maps to highlight tumor regions alongside their prediction confidence. Our experiments demonstrated that integrating an external dataset (Camelyon16) during training significantly improved model performance, surpassing the benefits of data augmentation alone. The optimal model, trained on both external and local data, achieved an accuracy and F1-score of 0.98, outperforming existing state-of-the-art methods.

Conclusions:

This study demonstrates that transfer learning architectures, when enhanced with multi-source data integration and interpretability frameworks, can significantly improve metastatic detection in whole slide imaging. Our methodology achieves diagnostic performance comparable to gold-standard techniques while dramatically accelerating analytical workflows. The synergistic combination of external dataset incorporation and probabilistic visualization outputs provides a clinically actionable solution that maintains both computational efficiency and pathological interpretability.