Unsupervised Out-of-Distribution Detection in Medical Imaging Using Multi-Exit Class Activation Maps and Feature Masking (MECAM-OOD)

Dataset

ID Dataset - ISIC19, PathMNIST

• ISIC 2019
• PathMNIST (from medmnist import PathMNIST)

DATASET_NAME
|-- train
|   |-- class_1
|   |   |-- xxx.jpg
|   |   |-- ...
|   |-- class_2
|   |-- class_3
|   |-- ...
|-- test
|   |-- class_1
|   |   |-- xxx.jpg
|   |   |-- ...
|   |-- class_2
|   |-- class_3
|   |-- ...

OOD Dataset - RSNA, COVID-19, HeadCT

• RSNA
• COVID-19
• HeadCT

DATASET_NAME
|-- class_1
|   |-- xxx.jpg
|   |-- ...
|-- class_2
|-- class_3
|-- ...

Train Classification Model with ID Dataset

cd CNet_ME
python3 train_cnet.py -b 128 -d ISIC --model_type Res18 -e 200

Run MECAM-OOD

cd MECAM-OODD
python3 main.py -id ISIC -ood RSNA --pretrained_path MERes18_ISIC_1231_003419 -gid 0