Multi-Organ-Segmentation-Using-Deep-Learning

NiftyNet framework based on Tensorfow was used for this project.

pip install niftynet

Dataset

Dataset was acquired from MICCAI 2012, MRBrains18 and ADNI.

Preprocessing

3D medical images were used as dataset and due to their large sizes and smaller number, they had to be resampled and augmented. SimpleITK library was used for preprocessing as seen in data_aug.pyfile.

pip install SimpleITK

Furthermore, data was increased by collecting it from various sites like MICCAI 2012, MRBRains18 and ADNI. Some images came with their annotations but some of them had to be segmented manually for training. ITKSNAP was used for manual annotations of 3D images.

Deep Learning Network

HighRes3D Networks were used for segmentation of 3 regions and 8 regions respectively. Checkpoints have been uploaded.

Results

Dice score of 0.957 and 0.776 was achieved for 3 brain regions and 8 brain regions respectively.