notesum.ai

Published at December 3

MERGE: Multi-faceted Hierarchical Graph-based GNN for Gene Expression Prediction from Whole Slide Histopathology Images

cs.CV

Released Date: December 3, 2024

Authors: Aniruddha Ganguly¹, Debolina Chatterjee², Wentao Huang¹, Jie Zhang², Alisa Yurovsky¹, Travis Steele Johnson², Chao Chen¹

Aff.: ¹Stony Brook University, NY, USA; ²Indiana University School of Medicine, IN, USA

Arxiv: http://arxiv.org/pdf/2412.02601v1

Refer to caption

Method	Graph	MSE $\downarrow$	MAE $\downarrow$	PCC $\uparrow$
ResNetSimCLR	-	$0.1999\pm 0.0164$	$0.3448\pm 0.0157$	$0.5221\pm 0.0465$
BLEEP	-	$0.3756\pm 0.0456$	$0.3756\pm 0.0456$	$0.0784\pm 0.0118$
TRIPLEX	-	$0.1472\pm 0.0206$	$0.2943\pm 0.0217$	$0.2320\pm 0.0199$
MERGE (w/ Multi-Res)	1-hop	$0.1482\pm 0.0181$	$0.2984\pm 0.0191$	$0.6458\pm 0.0478$
MERGE (w/ Multi-Res)	Hierarchical	$0.1461\pm 0.0186$	$0.2959\pm 0.0192$	$0.6566\pm 0.0474$
MERGE (w/ ResNet)	1-hop	$0.1783\pm 0.0133$	$0.3207\pm 0.0148$	$0.5789\pm 0.0360$
MERGE (w/ ResNet)	Hierarchical	$\mathbf{0.1347\pm 0.0164}$	$\mathbf{0.2834\pm 0.0183}$	$\mathbf{0.6795\pm 0.0386}$