3D Medical Image Segmentation

This project focuses on segmenting 3D MRI/CT scan images into various 15 organs. The primary architecture utilized is a custom 3D UNet, which is built upon the 2D UNet framework from 2015. Reference papers can be found in the "Papers" folder.

Training and Validation

To run training/validation, execute main.py. Adjust the patch size and batch size based on the available GPU size in your system. The model was trained on an NVIDIA TITAN RTX with 24GB of RAM.

Hyperparameters

Inside the config file, the following hyperparameters are defined:

• IMAGE_SIZE: Size of an image patch.
• PATCH_BATCH_SIZE: Number of patches provided to the GPU during training.
• INPUT_DIM: Number of channels in the input image (1 for MRI/CT, 3 for others).
• OUTPUT_CHANNEL: Number of output channels, i.e., the number of segments.

Loss Function

def dice_loss(input_im, target):
    smooth = 1.0
    iflat = input_im.flatten()
    tflat = target.flatten()
    intersection = (iflat * tflat).sum()

    return 1 - ((2. * intersection + smooth) / (iflat.sum() + tflat.sum() + smooth))

Other Approaches

While the main model is trained using image patches, other approaches can be explored using interpolation techniques (not recommended). Additional approaches can be found in the notebooks.