Auto Contouring of OAR in Pelvic CT Images Using an Encoder-Decoder Based Deep Residual Network

Automatic contouring of organs at risk is vital in radiotherapy treatment planning for curing any cancer disease. To calculate accurate distribution dose of radiation to the cancer cells requires the identification of nearby organs to safeguard them from irradiation. And it is very challenging in the pelvis area of the human body due to unclear boundaries, bowel gas, and the varying size of organs in different people. Deep learning-based automatic contouring can mitigate manual contouring difficulties. In this paper, a modified U-Net with a particular residual network(ResNext), having a horizontal dimension in addition to depth and width, is used for contouring of organs, namely, bladder and left & right femoral heads on pelvic Computed Tomography(CT) images. The results reveal that the dice coefficient value obtained from the experiments is above 0.88 for all three organs. The model outperformed the typical U-Net and is comparable to other state-of-the-art models in this area. The proposed model can be used for automatic contouring of pelvic organs and reduce treatment planning time significantly.