Automatic Liver Tumor Segmentation
The liver is a common site of primary or secondary tumor development. Due to their heterogeneous and diffusive shape, automatic segmentation of tumor lesions is very challenging. We developed an automatic segmentation algorithms to segment liver lesions in contrastÂ-enhanced abdominal CT scans.
Our problem statement is inspired by a challenge on CodaLabs. The training data set contains 130 CT scans, entailing segmentation masks corresponding to each scan.
Approach
We have used Unet, which is a convolutional network architecture for fast and precise segmentation of images: An 8-layer deep encoder and decoder, with weighted binary cross entropy loss function, and dice-coefficient as evaluation metric. The model performs extremely well with a 0.83 dice-coefficient on validation set, after 20 epochs of training.
The demo and network is explained in the following video:
Upload your CT Scan [Functional only with a server]
The interim hackathon proposal can be found here and the code can be found here.