Repository for the 2017 Digital Mammography DREAM Challenge
See the installation instruction for a step-by-step installation guide. See the server instruction for server setup.
- Install Torch
- Install cuda-8.0
- Install cudnn v5.1
- Install luarocks packages
$ luarocks install cutorch
$ luarocks install xlua
$ luarocks install optnet- Calcification-Classifier : Classification whether the ROI region of calcification is malignant/benign.
- Mass-Classifier : Classification whether the ROI region of mass is malignant/benign.
- ornot-Calcification : Deciding whether the given window is a calcification ROI.
- ornot-Mass : Deciding whether the given window is a mass ROI.
The input of the calcification-classifier will be a square window of the ROI region of 'calcification'. ROI regions will be extracted according to the heatmap derived from ornot_calcification.
- Input size : 256 x 256
- Crop size : 224 x 224
- Model : Fine-tuned Residual Network 50 (ILSVRC-2012)
- Best acc : 80%
The input of the mass-classifier will be a square window of the ROI region of 'mass'. ROI regions will be extracted according to
Total score = (Faster-RCNN results) + (Distance comparison of CC, MLO views) + ([ornot-Mass](./ornot-Mass/) results)of the 'mass' regions in our private dataset.
- Input size : 256 x 256
- Crop size : 224 x 224
- Model : Fine-tuned Residual Network 50 (ILSVRC-2012)
- Best acc : None
The input of the ornot-calcification will be a tiny window of the suspected region of 'calcification'. Regions will be extracted according to Faster-RCNN training of the 'calcification' regions in our private dataset.
- Input size : 36 x 36
- Crop size : 32 x 32
- Model : Wide-Residual-Network 28x10
- Best acc : 96%
The input of the ornot-mass will be a square window of the suspected region of 'mass'. Regions will be extracted according to Faster-RCNN training of the 'Mass' regions in our private dataset.
- Input size : 256 x 256
- Crop size : 224 x 224
- Model : Fine-tuned Residual Network 50 (ILSVRC-2012)
- Best acc : 83%