Data preparation script is used to convert the raw data to convert from MHA Archive to nnU-Net Raw Data Archive and also split the data into splits. This script is provided by Diagnostic Image Analysis Group for this challenge.
All the train, valid data-splits are created by plan_overview.py script, also provided by the Diagnostic Image Analysis Group.
SimpleITK library is used to read and process the medical images for detection.
We used the monai framework to create the Unet model. The loss function and metrics are provided by the Grand Challenge.
- Focal Loss (Binary Segmentation loss)
- Average Precision (AP)
- Area Under the Receiver Operating Characteristic curve (AUROC)
- Overall AI Ranking Metric of the PI-CAI challenge: (AUROC + AP) / 2
- Precision-Recall (PR) curve
- Receiver Operating Characteristic (ROC) curve
- Free-Response Receiver Operating Characteristic (FROC) curve