MedAI- Graph-Based Modeling in Computational Pathology - Siyi Tang

Explore graph-based modeling techniques in computational pathology through this comprehensive lecture by Stanford University PhD candidate Siyi Tang. Delve into the emerging field of leveraging cellular interactions and spatial structures in whole slide images using graph neural networks. Learn about spectral and spatial networks, cell graph construction, adaptive glossage, node embedding, and graph clustering. Examine experiments in cluster assignments, cancer grading, and nuclear sampling. Analyze drawbacks, post-hoc graph expanders, and evaluation metrics like separability scores. Gain insights into qualitative and quantitative assessments, as well as personal takeaways on domain expertise and explanation methods. Engage with cutting-edge research aimed at developing better medical machine learning models and enabling novel scientific discoveries in pathology.

Introduction
Outline
Spectral networks
Spatial networks
Computational pathology
Cell graphs
Cell graph convolutional network
Method overview
Cell graph construction
What is graph adaptive glossage
What is adaptive glossage
Node embedding
Graph clustering
Graph class
Concatenation
Experiments
Cluster assignments
Cluster interpretation
Cancer grading
Nuclear sampling
Overview
Graph construction
Graph new network
Quick question
Paper
Drawbacks
Posthoc graph expanders
Histograms
Separability Score
Aggregate
Final Score
Risk Score
Data Set
Qualitative Assessment
Quantitative Assessment
Personal takeaways
Domain expertise
Explanation explainers
Graph neighborhood sampling
Questions