NLP to Extract Data from Cancer Pathology Reports with Human-Level Accuracy
, Moffitt Cancer Center
Highly Rated
Learn how we achieved human-level performance extracting cancer data from pathology reports. We'll explain BERT and how we used it to predict International Classification of Diseases, Oncology, version 3.2 (ICD-O-3) tumor-site and histology codes. We'll explain why this task is so important, and its multiple uses within cancer care. Then we'll present new GPU solutions to this longstanding problem, and finally discuss future efforts and applications.
To be covered: GPU NLP techniques including BERT and its health-care-specific variants. Cancer pathology, with specific and interesting tumor site and histology examples. High-level descriptions of our system architecture and the methods to train and test it. You'll benefit from having some prior knowledge of NLP. Familiarity with cancer treatment and research will help you judge the relevance and applications of this work. Specific programming and implementation knowledge aren't required.