Multi-Organ Auto-segmentation using Deep Learning approaches for head and neck section for Cancer treatment

Cancer treatment using radiotherapy has been drawing the attention of biomedical researchers. Radiotherapy treatment requires specialists to contour the tumor and organs at risk (OARs) which is a time-consuming task. Manual contouring is repetitious and vulnerable to human mistakes. This leads to a decrease in the number of patients treated daily. The previously stated facts have motivated researchers to channel their efforts into auto-segmentation methods for the automation of tumor contouring. Auto-segmentation architectures play a vital role in reducing the enormous workload faced by medical personnel. In our case, we developed our proposed Res-U-Net which is based on the U-Net architecture. The used dataset was the MICCAI dataset which consists of 9 organ's CTs each having its own ground truth. In our approach, we concatenated the organ's masks altogether making it a multiclass segmentation problem. Our proposed Res-U-Net achieved 95.68% dice coefficient (DSC).