GANs in Medical Image Synthesis, Translation, and Augmentation - Jason Jeong

Explore the applications of Generative Adversarial Networks (GANs) in medical imaging through this comprehensive lecture by Jason Jeong from Stanford University. Delve into the use of GANs for medical image synthesis, translation, and augmentation, with a focus on addressing data scarcity and imbalance in healthcare datasets. Learn about the implementation of GANs in various imaging modalities such as CT, MRI, Ultrasound, and PET for improved disease diagnosis and assessment. Discover the speaker's recent work on generating synthetic dual energy CT from single energy CT, and gain insights into the challenges and future directions of GANs in medical imaging. The lecture covers GAN architecture, medical applications in classification and segmentation, distribution of GANs, and specific case studies. Engage with topics like improving intracranial image detection, data balance problems, conditional data augmentation, and common GAN issues such as mode collapse. This 56-minute talk is part of the MedAI Group Exchange Sessions, offering a platform for critical examination of AI and medicine intersections.

Introduction
Overview
GAN Architecture
Medical GANs
Classification and Segmentation
Distribution of GANs
Applications of GANs
Study
Translation Results
External Results
Numerical Results
External Data
Numerical Data
Improving Intracranial Image Detection
Data Balance Problem
Classification Method
Binary Classification
Conditional Dance Augmentation
Confusion Matrix
Epidural Cases
Mode Collapse
GAN Problems
Fenscan
Decision Boundary
Boundary Decision Guns
Questions