Model Card for MuCTaL

Model Details

Model Description

Multi-cancer tile classifier. Predict tumor / not-tumor from 224px H&E stain normalized tiles. Ues acral MEL, HCC, Lung and CRC Lung: Kaggle. CRC: [Zenodo] (https://zenodo.org/records/1214456)

• Developed by: Brian Isett, University of Pittsburgh
• Model type: Densenet169
• Finetuned from model [optional]: Densenet169 IMAGENET1K_V1 via fastai / pytorch

Model Sources [optional]

• Repository: https://github.com/AivaraX-AI/MuCTaL
• Paper: [Fothcoming]

Uses

Direct Use

from fastai.vision.all import *
model_fn = 'full_model.pkl'
learn = load_learner(model_fn, cpu = False)
dl = learn.dls.test_dl(valid_df.loc[:,'fn'] ) #valid_df is dataframe of 224 x 224 px H&E tile file names ('fn')
pred = learn.get_preds(dl=dl, with_decoded=True)

Downstream Use [optional]

https://github.com/HCC-data-sciences/pancancer_he_classifier/blob/main/06_acral_tile_heatmap_class_viz.ipynb

Out-of-Scope Use

This model is for research use only. Not to be used in a clinical or patient diagnostic setting.