Project 466810

The standard of care for locally advanced cervical cancer includes external beam radiation therapy (EBRT) with concomitant chemotherapy, followed by brachytherapy (BT). BT involves the positioning of a small radioactive source in or close to a tumor using needles or catheters, and has the benefit of a higher radiation dose to the tumor with less exposure to nearby healthy tissues compared to EBRT. One drawback to BT is that the procedure is time intensive. Once the applicator and needles have been inserted, it is necessary to obtain CT or, preferably, MR imaging of the patient in order to confirm the location of the channels relative to anatomical features. Currently, clinical target volumes (CTVs) and nearby healthy organs-at-risk (OARs), as well as the BT applicator, must be manually delineated by a clinician and incorporated into treatment planning. This delineation often takes between 45 and 90 minutes, and thus is a viable target for automation to minimize the total time required for a BT procedure.The aim of the proposed work is to construct a deep learning (DL) model capable of adequately segmenting volumetric MR images for use in BT treatment planning. The model will be trained and tested on previously collected cervical cancer volumetric MRI datasets and their corresponding clinical segmentations. The benefits of automating OARs and CTVs delineation include decreased organ filling between imaging and treatment, decreased time for which the patient is either bedridden or under anaesthesia, and decreased inter-observer segmentation variability, all of which can improve treatment outcomes. This work endeavours to improve the accuracy and convenience of MRI-guided BT for cervical cancer.