The patches are cropped from the images of DDSM (Digital Database for Screening Mammography) dataset. A regular way to formulate the problem is to first detect the region of interest, and then classify into malignant, and non-malignant. Each of the folders of Training and Validation contains Excel Sheets to indicate the location of the patches (x-min, y-min, x-max, y-max) under consideration. The dataset contains 1855 malignant patches for training and 625 for validation. It contains 3743 non-malignant patches for training and 1232 for validation.
Three models were built -
- Resnet Model
- Efficient Model
- Custom CNN Model
The Model Weights can be accessed from my google drive after a valid permission to use - https://drive.google.com/drive/folders/1T1d67WnpECmrZ5cdTpJAzHm8Lz7WZZ2h Mail me at [email protected] for any doubts.