Histoboard - Pathology Foundation Model Benchmarks

HEST-Benchmark
spatial-transcriptomics
H&E

HEST-Benchmark is a spatial transcriptomics benchmark from the Mahmood Lab evaluating foundation models on gene expression prediction from H&E images. The benchmark assesses how well models can predict spatially-resolved gene expression patterns directly from histology across 9 cancer types using Pearson correlation as the primary metric.

25 models evaluated

9 tasks

Organs:

breast

prostate

pancreas

skin

colon

rectum

kidney

lung

lymph

Detailed Results

Model	Average rank	Average metric	ccRCC (Kidney) Pearson	COAD (Colon) Pearson	IDC (Breast) Pearson	LUAD (Lung) Pearson	Lymph IDC Pearson	PAAD (Pancreas) Pearson	PRAD (Prostate) Pearson	READ (Rectum) Pearson	SKCM (Skin) Pearson
1H-optimus-1	2.44	0.423	0.253	0.320	0.602	0.578	0.277	0.496	0.378	0.242	0.659
2H-optimus-0	3.44	0.415	0.268	0.309	0.598	0.559	0.259	0.491	0.385	0.222	0.645
3UNI2	4.00	0.414	0.264	0.301	0.590	0.559	0.273	0.500	0.357	0.222	0.661
4Virchow	6.67	0.406	0.259	0.308	0.585	0.566	0.261	0.516	0.338	0.198	0.624
5Virchow2	6.56	0.403	0.272	0.258	0.597	0.569	0.257	0.478	0.353	0.207	0.640
6H0-mini	7.67	0.396	0.267	0.249	0.586	0.548	0.263	0.492	0.369	0.186	0.601
7Midnight-12k	9.89	0.395	0.213	0.291	0.582	0.558	0.264	0.490	0.337	0.186	0.636
8OpenMidnight	8.89	0.391	0.246	0.273	0.587	0.553	0.260	0.473	0.359	0.176	0.594
9Hibou-L	10.00	0.388	0.266	0.304	0.570	0.576	0.243	0.467	0.294	0.190	0.582
10Prov-GigaPath	10.67	0.387	0.243	0.299	0.551	0.541	0.250	0.475	0.370	0.196	0.562
11UNI	11.44	0.387	0.240	0.261	0.589	0.546	0.256	0.481	0.294	0.184	0.635
12GPFM	13.56	0.379	0.259	0.248	0.566	0.547	0.237	0.460	0.342	0.165	0.589
13CONCH 1.5	12.33	0.379	0.218	0.280	0.544	0.551	0.270	0.457	0.381	0.160	0.552
14Phikon-v2	13.89	0.375	0.266	0.250	0.541	0.542	0.244	0.446	0.354	0.175	0.555
15Kaiko ViT-B/8	14.44	0.373	0.231	0.268	0.560	0.518	0.227	0.460	0.361	0.162	0.573
16CONCH	15.89	0.370	0.218	0.249	0.536	0.532	0.251	0.447	0.355	0.160	0.579
17Lunit ViT-S/8	15.22	0.368	0.246	0.283	0.545	0.541	0.251	0.427	0.283	0.161	0.574
18Phikon	16.89	0.366	0.242	0.262	0.533	0.547	0.237	0.443	0.342	0.153	0.535
19Kaiko ViT-B/16	18.00	0.364	0.229	0.281	0.535	0.516	0.236	0.452	0.328	0.152	0.550
20Kaiko ViT-L/14	17.78	0.364	0.219	0.254	0.553	0.538	0.228	0.437	0.347	0.147	0.553
21Kaiko ViT-S/8	20.89	0.351	0.235	0.228	0.530	0.505	0.246	0.418	0.334	0.147	0.517
22Kaiko ViT-S/16	21.33	0.349	0.210	0.206	0.533	0.503	0.225	0.441	0.348	0.133	0.545
23CTransPath	19.89	0.347	0.236	0.238	0.499	0.514	0.241	0.431	0.355	0.097	0.510
24MUSK	20.44	0.347	0.182	0.236	0.525	0.517	0.255	0.428	0.343	0.111	0.523
25ResNet50 (ImageNet)	22.78	0.325	0.225	0.250	0.474	0.495	0.231	0.388	0.304	0.078	0.482

HEST-Benchmarkspatial-transcriptomicsH&E

Detailed Results

HEST-Benchmark
spatial-transcriptomics
H&E