ODySSeI

An Open-Source End-to-End Framework for Automated Detection, Segmentation, and Severity Estimation of Lesions in Invasive Coronary Angiography Images

Web Interface

ODySSeI is live at swisscardia.epfl.ch.

Please find below a demo of ODySSeI's web interface. Demo images are available in the demo_images folder within the data folder.

ODySSeI_Demo_Final.mp4

Setting Up the Repository

To use ODySSeI, please run the following code snippet in your terminal:

git clone https://github.com/LTS4/ODySSeI
cd odyssei
python -m pip install -e .

Next, please place your custom ICA dataset folder, custom_dataset, within the data folder.

Training, Validation, and Testing of our Lesion Detection Model (YOLO11m)

Please run the following code snippet in your terminal:

python odyssei/lesion_detection_train_val_test.py --pretrained_model_file=PRETRAINED_MODEL_FILE --custom_dataset=CUSTOM_DATASET --wandb_project=WANDB_PROJECT --wand_run=WANDB_RUN --num_epochs=NUM_EPOCHS

Here,

• PRETRAINED_MODEL_FILE (str) = Pretrained Model Weights (yolo11m.pt)
• CUSTOM_DATASET (str) = Name of the custom_dataset folder in the data folder; Please follow the YOLO dataset format
• WANDB_PROJECT (str) = Name of the W&B Project
• WANDB_RUN (int) = Name of the W&B Run for Logging Results
• NUM_EPOCHS (int) = Number of Training Epochs

Training, Validation, and Testing of Lesion Segmentation Models (DeepLabv3+ and U-Net)

Please run the following code snippet in your terminal:

For DeepLabv3+:

python odyssei/deeplab_lesion_segmentation_train_val_test.py --pretrained_model_file=PRETRAINED_MODEL_FILE --custom_dataset=CUSTOM_DATASET --wandb_project=WANDB_PROJECT --wand_run=WANDB_RUN --num_epochs=NUM_EPOCHS

For U-Net:

python odyssei/unet_lesion_segmentation_train_val_test.py --custom_dataset=CUSTOM_DATASET --wandb_project=WANDB_PROJECT --wand_run=WANDB_RUN --seed=SEED --batch_size=BATCH_SIZE --num_epochs=NUM_EPOCHS --loss_func=LOSS_FUNC

Here,

• CUSTOM_DATASET (str) = Name of the custom_dataset folder in the data folder; Please follow the following structure:

  • CUSTOM_DATASET (Name of the Custom Dataset)
    |- images (This is the subfolder where you need to save your ICA images)
    |- masks (This is the subfolder where you need to save your corresponding ground truth segmentation masks)
    

• WANDB_PROJECT (str) = Name of the W&B Project
• WANDB_RUN (int) = Name of the W&B Run for Logging Results
• SEED (int) = Seed for Reproducibility
• BATCH_SIZE (int) = Batch Size (Recommended: 16)
• NUM_EPOCHS (int) = Number of Training Epochs
• LOSS_FUNC (int) = Loss Function, e.g., bce1, bce2, bce3, iou, dice, cldice, dicecldice, dicebce1, dicebce2 (Recommended: dicebce2)