From: Pathology Foundation Models
| Characteristics | Foundation model | Nonfoundation model (deep learning model) |
|---|---|---|
| Model architecture | (Mainly) Transformer | Convolutional neural network |
| Model size | Vary large | Medium to large |
| Model applicable task | Many | Single |
| Performance on adapted tasks | State-of-the-art (SOTA) | High to SOTA |
| Performance on untrained tasks | Medium to high | Low |
| Data amount for model training | Very large | Medium to large |
| Using labeled data for model training? | No | Yes |
From: Pathology Foundation Models
| Foundation model name | CTransPath | (Lunit)* | PLIP | Virchow | UNI | CONCH | PRISM | Prov-GigaPath | TANGLE | RudolphV |
|---|---|---|---|---|---|---|---|---|---|---|
| Publication date | 2022 | 2022 | 2023 | 2024 | 2024 | 2024 | 2024 | 2024 | 2024 | 2024 |
| Journal name | Medical Image Analysis | arxiv | Nature Medicine | Nature Medicine | Nature Medicine | Nature Medicine | arxiv | Nature | CVPR2024 | arxiv |
| Pretraining dataset name | TCGA | TCGA | OpenPath | MSKCC | Mass-100K | Educational sources | MSKCC) | Dataset from Providence | TCGA | Dataset from over 15 different laboratories across the EU and US |
| PAIP | TULIP | Mass-1K | PubMed Central Open Access Dataset | TG-GATEs | TCGA | |||||
| Mass-22K | ||||||||||
| Number of GPUs/type used for training | 48/NVIDIA V100 GPUs | 64/NVIDIA V100 GPUs | Not specified | -/NVIDIA A100 GPUs | 32/NVIDIA A100 GPUs | 8/NVIDIA A100 GPUs | 16/NVIDIA V100 GPUs | 16 nodes × 4/NVIDIA A100 GPUs | 8/NVIDIA A100 GPUs | 16/NVIDIA A100 GPUs |
| Embedding level (patch/slide) | Patch | Patch | Patch | Patch | Patch | Patch | Slide | Slide | Slide | Patch |
| Number of WSIs | 29,763 (TCGA) | 20,994 (TCGA) | Not specified | 1,488,550 | 100,426 | Not specified | 587,196 | 171,189 | 2,074 (TCGA) | 133,998 |
| 2,457 (PAIP) | 15,672 (TULIP) | 6,597 (TG-GATEs) | ||||||||
| Number of patch images (M) | 15 | 32.6 | 0 | 2,000 | 100 | 1 | Not specified | 1,300 | 15 | 1,200 |
| Number of patients | Not specified | Not specified | Not specified | 119,629 | Not specified | Not specified | 195,344 | >30,000 | >1,864 | 34,103 |
| More than 10 types of organs in the dataset? | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | No | Yes |
| Staining types (H&E/H&E + others) | Not specified | H&E | H&E + others | H&E | H&E | H&E + others | H&E | H&E + others | Not specified | H&E + others |
| FFPE/frozen | FFPE, frozen | Not specified | Not specified | FFPE | FFPE | Not specified | Not specified | Not specified | Not specified | FFPE, Frozen |
| VFM/MFM | VFM | VFM | MFM | VFM | VFM | MFM | MFM | MFM | MFM | VFM |
| Model publicly available? | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | No | Yes |
| Reference number | 33 | 25 | 28 | 27 | 28 | 30 | 31 | 22 | 10 | 32 |
