Predicting breast cancer metastasis in sentinel lymph nodes with artificial intelligence
Background. Deep learning, a specialized technique in artificial intelligence, has been shown to be useful to detect breast cancer metastases from analyzing whole slide image (WSI) of sentinel lymph nodes, however, it requires extensive scanning and analysis of all the lymph node slides for each case. Our deep learning study focuses on breast cancer screening with only a small set of image patches from any sentinel lymph node to detect changes in tumor environment and not in the tumor itself.
Methods. This study involves breast pathologists in our department and uses our in-house breast cancer cases and WSI scanners. We design a convolutional neural network in Python language, together with TensorFlow and Keras libraries, to build a diagnostic model for four diagnostic categories (macro metastasis, micrometastasis, isolated tumor cells, and negative metastasis). For each test case, the predicted diagnosis is combined from the prediction for 5 images (at least 3 or more must agree), a process known as “majority voting”. We obtained WSIs of Hematoxylin and Eosin stained slides from 34 cases with approximate equal distribution in 4 diagnostic categories. A positive WSI and a negative WSI were elected for each case to obtain a total of 68 WSIs. From each WSI, 40 image patches (100x100 pixels) were obtained to yield 2720 image patches, from which 2160 (79%) were used for training, 240 (9%) for validation, and 320 (12%) for testing.
Results. The test results showed excellent diagnostic results: accuracy (91.15%), sensitivity (77.92%), specificity (92.09%), positive predictive value (90.86%), and negative predictive value (80.66).
Conclusions. This preliminary study provided a proof of concept for incorporating automated metastatic screen into the digital pathology workflow to augment the pathologists’ productivity. Our approach is unique since it provides a very rapid screen rather than an exhaustive search for tumor in all fields of all sentinel lymph nodes.
Objectives:
- Understand how deep learning screens for breast cancer metastasis using whole slide images of sentinel lymph nodes
- Know the difference between diagnostic criteria based on histology in tumor (requiring all sentinel lymph nodes) and that based on tumor environment (requiring a small subset of sentinel lymph nodes)
- Know how screen results by deep learning can improve microscopic diagnosis
Presented by:
Nghia Nguyen, MD
Vice Chair, Digital Pathology
Professor of Pathology
University of Texas-Houston Medical School
Andy Nguyen received his BS and MS degrees in Mechanical Engineering from the University of Houston. He obtained his MD degree from the University of Texas Medical Branch-Galveston and his pathology residency training at the University of Texas Health Sciences Center, Houston. He served as medical director of hematopathology (1993-2021) in the same department. He is currently Vice Chair, Digital Pathology. With extensive background in computer programming, he has developed decision-support software modules covering topics in hematology, coagulation, and molecular hematopathology (http://hemepathreview.com). Published research findings in benign/malignant hematologic disorders, coagulopathy, hematologic mutations, and pathology informatics. Chaired the whole slide imaging (WSI) task group to select WSI scanners to allow pathologists to sign out pathology cases using digital images. His current research interests lie in designing deep learning software for screening and diagnosis of pathologic lesions using WSI from various organ systems including lymphoma and carcinoma.
