Curation and Analysis of Annotated Medical Images Across Institutions - Xiaoyuan Guo

Explore a comprehensive lecture on facilitating the curation and analysis of annotated medical images across institutions. Delve into the challenges of data insufficiency in supervised deep learning approaches for medical imaging tasks and discover innovative solutions. Learn about unsupervised anomaly detection techniques for identifying in-distribution and out-of-distribution data, as well as methods for quantifying dataset quality. Examine a novel content-based medical image retrieval method that balances intra- and inter-class variance for OOD-sensitive retrieval. Gain insights into accelerating the curation process through automatic detection of noisy and under-represented data. Understand the potential applications of these techniques in image annotation, querying, and future analysis of external datasets. Presented by Xiaoyuan Guo, a Computer Science PhD student at Emory University, this talk covers key topics including deep learning approaches, shift data categories, model exchange pipelines, and experimental results in the context of medical image processing and computer vision.

Introduction
Outline
Applications
Deep approaches
Data insufficiency
Shift data examples
Shift data categories
Challenges
Model Exchange
Pipeline
Representation
Architecture
Intermediate Results
Clustering
Elbow Method
Classification
Portal
Shift Data Quantification
Outlier Sensitive ContentBased Image Retrieval
Pseudolabel
Enterprise Differences
Experimental Results
Hand Example
Discussion