Poster: Head and Neck Tumor Segmentation With Sliced 3D PET Scans

We present a semi-automatic 3D segmentation technique for head and neck tumors in PET Scans. First, we slice the 3D PET Scans into 2D images and separate the slices into slices with tumor and slices without tumor. The slices with tumor get segmented by a 2D U-Net. After training the U-Net for 100 epochs, the model achieved a dice score of 0.7799 and a dice loss of 0.3732.