Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning for Identifying Metastatic Breast Cancer

Dayong Wang, Aditya Khosla, Rishab Gargeya, Humayun Irshad, Andrew H. Beck

TL;DR

The paper tackles automated detection of metastatic breast cancer in whole-slide sentinel lymph node images via a patch-based deep CNN that generates tumor heatmaps. By post-processing these heatmaps, the authors derive slide-level and lesion-level predictions, achieving top performance on Camelyon16 with a slide-level AUC of 0.925 and lesion detection score of 0.7051. Importantly, integrating the deep learning system with a pathologist markedly improves diagnostic accuracy (AUC ~0.995) and reduces human error, illustrating the practical value of AI-assisted pathology. The study demonstrates how large-scale patch training, heatmap aggregation, and carefully designed post-processing can push pathological diagnostics toward near-human performance while reducing cognitive load.

Abstract

The International Symposium on Biomedical Imaging (ISBI) held a grand challenge to evaluate computational systems for the automated detection of metastatic breast cancer in whole slide images of sentinel lymph node biopsies. Our team won both competitions in the grand challenge, obtaining an area under the receiver operating curve (AUC) of 0.925 for the task of whole slide image classification and a score of 0.7051 for the tumor localization task. A pathologist independently reviewed the same images, obtaining a whole slide image classification AUC of 0.966 and a tumor localization score of 0.733. Combining our deep learning system's predictions with the human pathologist's diagnoses increased the pathologist's AUC to 0.995, representing an approximately 85 percent reduction in human error rate. These results demonstrate the power of using deep learning to produce significant improvements in the accuracy of pathological diagnoses.

Deep Learning for Identifying Metastatic Breast Cancer

TL;DR

The paper tackles automated detection of metastatic breast cancer in whole-slide sentinel lymph node images via a patch-based deep CNN that generates tumor heatmaps. By post-processing these heatmaps, the authors derive slide-level and lesion-level predictions, achieving top performance on Camelyon16 with a slide-level AUC of 0.925 and lesion detection score of 0.7051. Importantly, integrating the deep learning system with a pathologist markedly improves diagnostic accuracy (AUC ~0.995) and reduces human error, illustrating the practical value of AI-assisted pathology. The study demonstrates how large-scale patch training, heatmap aggregation, and carefully designed post-processing can push pathological diagnostics toward near-human performance while reducing cognitive load.

Abstract

The International Symposium on Biomedical Imaging (ISBI) held a grand challenge to evaluate computational systems for the automated detection of metastatic breast cancer in whole slide images of sentinel lymph node biopsies. Our team won both competitions in the grand challenge, obtaining an area under the receiver operating curve (AUC) of 0.925 for the task of whole slide image classification and a score of 0.7051 for the tumor localization task. A pathologist independently reviewed the same images, obtaining a whole slide image classification AUC of 0.966 and a tumor localization score of 0.733. Combining our deep learning system's predictions with the human pathologist's diagnoses increased the pathologist's AUC to 0.995, representing an approximately 85 percent reduction in human error rate. These results demonstrate the power of using deep learning to produce significant improvements in the accuracy of pathological diagnoses.

Paper Structure

This paper contains 15 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Dataset and Evaluation Metrics
  3. Camelyon16 Dataset
  4. Evaluation Metrics
  5. Method
  6. Image Pre-processing
  7. Cancer Metastasis Detection Framework
  8. Patch-based Classification Stage
  9. Post-processing of tumor heatmaps to compute slide-based and lesion-based probabilities
  10. Lesion-based Detection
  11. Experimental Results
  12. Evaluation Results from Camelyon16
  13. Combining Deep Learning System with a Human Pathologist
  14. Discussion
  15. Acknowledgments

Figures (5)

  • Figure 1: Visualization of tissue region detection during image pre-processing (described in Section \ref{['sec:preprocessing']}). Detected tissue regions are highlighted with the green curves.
  • Figure 2: The framework of cancer metastases detection.
  • Figure 3: Visualization of tumor region detection.
  • Figure 4: Receiver Operating Characteristic (ROC) curve of Slide-based Classification
  • Figure 5: Free-Response Receiver Operating Characteristic (FROC) curve of the Lesion-based Detection.