Mandible Segmentation from Head and Neck CT Volumes

Data

Download the head and neck data from the MICCAI Challenge site.

Data preparation

Data split has been listed in dataList.ods.

Patch and Slice extraction codes inside HeadNeck/codes

Training

Trains a patch based CNN to classify whether the given patch belongs to Mandible or not. Patch size of 21x21 was chosen.

TODO:

Convert the patch based network into a fully convolutional network.

Results

The best network was tested on Validation+Testing Dataset and the dice similarity score was 0.53.