Automated volumetric delineation of cancer tumors on PET/CT images using 3D convolutional neural network (V-Net)

Abstract

The process of delineation of tumors and malignant lymph nodes using medical images is a fundamental part of radiotherapy planning. Still, this process is done manually by radiologists. This process is time-consuming and suffers from inter-observer variability. Hence, there is a need to fully automate this process of delineation to reduce time consumption and inter-observer variability. Deep learning is a division of artificial intelligence that has proven to be useful for the automatic segmentation of medical images with the use of neural networks. We have to follow a systematic procedure as these neural networks require a large number of parameters to be tuned during the delineation process, to guarantee reproducibility. This thesis will present a complete theory of deep learning and a convolutional neural network, for delineating 3D images. The project will use the $deoxys$ framework to implement the V-Net architecture for automatic delineation of gross tumor volume and malignant lymph nodes in the head and neck region.