Explore a cutting-edge approach to improving machine learning model generalization in medical imaging through a 52-minute conference talk by Rikiya Yamashita from Stanford University. Delve into STRAP (Style TRansfer Augmentation for histoPathology), a novel data augmentation technique that uses non-medical artistic paintings to create domain-agnostic visual representations in computational pathology. Learn how this method enhances model robustness to domain shifts and achieves state-of-the-art performance in pathology classification tasks. Gain insights into the challenges of applying machine learning to medical imaging and discover potential solutions for improving clinical applicability. Understand the speaker's unique perspective as a radiologist turned applied research scientist and how this dual expertise contributes to bridging the gap between machine learning and clinical medicine.
Style Transfer Augmentations for Computational Pathology - Rikiya Yamashita
Stanford University via YouTube
Overview
Syllabus
Introduction
Rikiya Introduction
Computational Pathology
Batch Effect
Motivation
Domain adaptation
Single domain generalization
Like a
Texture bias
Recap
Method
Classification
Results
Style Transfer
Comparison
stylization coefficient
test data set
another paper
fastfree transformation
salience maps
identification
model performance
future work
texture vs shape bias
Yaxis
Questions
Discussion
Taught by
Stanford MedAI
