Histoboard - Pathology Foundation Model Benchmarks

PathoROB
pathology
robustness
H&E

Robustness benchmark evaluating pathology foundation models across domain shift scenarios including TCGA 2x2 splits, Camelyon, and Tolkach ESCA datasets.

23 models evaluated

3 tasks

Organs:

multi-organ

breast

esophagus

Detailed Results

Showing 3 domain shift scenarios. Robustness Index values - higher values indicate better robustness to distribution shifts.

Model	Average rank	Average metric	Camelyon Rob. Index	TCGA 2x2 Rob. Index	Tolkach ESCA Rob. Index
1Atlas 2^[1]	1.00	0.928	0.940	0.879	0.964
2GenBio-PathFM^[2]	2.33	0.888	0.865	0.838	0.960
3Virchow2	4.33	0.861	0.806	0.822	0.955
4CONCH 1.5	4.33	0.852	0.774	0.832	0.951
5Atlas	5.33	0.850	0.785	0.826	0.938
6Virchow	8.67	0.815	0.751	0.761	0.932
7H0-mini	8.67	0.815	0.718	0.794	0.932
8H-optimus-1^[2]	6.33	0.814	0.645	0.853	0.944
9CONCH	6.67	0.812	0.662	0.824	0.951
10H-optimus-0	9.33	0.812	0.705	0.812	0.918
11UNI2	10.67	0.757	0.544	0.803	0.923
12MUSK	12.67	0.707	0.467	0.727	0.928
13HIPT	17.00	0.663	0.649	0.614	0.726
14Prov-GigaPath	16.00	0.630	0.399	0.738	0.754
15Kaiko ViT-B/8	14.00	0.602	0.147	0.763	0.896
16UNI	14.67	0.598	0.145	0.747	0.902
17RetCCL	17.00	0.596	0.318	0.593	0.878
18CTransPath	17.67	0.543	0.106	0.652	0.872
19Lunit ViT-S/16	18.00	0.512	0.043	0.661	0.832
20RudolfV	20.00	0.489	0.184	0.587	0.695
21Phikon	19.67	0.476	0.011	0.623	0.795
22Phikon-v2	20.00	0.469	0.019	0.619	0.768
23Ciga	21.67	0.446	0.135	0.511	0.693

[1] Results taken from Atlas 2 preprint

[2] Results taken from GenBio-PathFM preprint

PathoROBpathologyrobustnessH&E

Detailed Results

PathoROB
pathology
robustness
H&E