Statistical Challenges in the Design and Validation of Medical Deep Learning Models

• Activity Number: 378 - Study Design and Statistical Challenges for AI/ML Based Medical Tests
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 12:00 PM to 1:50 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Andrew Beam*
• Companies: Harvard
• Keywords: deep learning; medical imaging ; artificial intelligence

Abstract

Diagnostic medical imaging has been transformed in recent years by deep learning models trained on large amounts of patient data. Though there has been much progress and enthusiasm, these approaches have shed new light on classic issues in confounding, study design, and model validation. In this talk, we will highlight key challenges in the deep learning model validation pipeline and provide examples of subtle ways that these challenges may present themselves at different stages of model development.