Automatic Segmentation of 3D MRI for Cervical Cancer Radiation Therapy with Transfer Learning

Sze-Nung Leung, Wei Dai, Jason Dowling, Karen Lim, Tony Young, Lois Holloway, Shekhar S. Chandra

Keywords: 3D multi-organ segmentation, MRI, Cervical Cancer, CNN, transfer learning

TL;DR: Automatic segmentation model on 3D Cervical Cancer MRI with transfer learning

Abstract: Cervical cancer is the fourth most common cancer in women worldwide. Treatment of the disease includes radiation therapy treatment, requiring precise organ delineation for treatment planning. Improvement of the segmentation process for the tumour site and nearby organs could limit treatment side effects and improve treatment planning efficiency. Compared with segmentation for other anatomical locations, Magnetic Resonance Imaging (MRI) for cervical cancer is challenging based on the limited amount of training data and large inter-patient shape variation for organs close to the tumour (organs-at-risk). In this paper, automatic 3D segmentation networks were studied with transfer learning, targeting 3D cervical cancer MRI data. The approaches were compared to the 3D U-Net which was widely used in recent studies in image segmentation for cervical cancer MRI or computed tomography (CT). The data used consisted of 44 images obtained from 20 patients with stage IB to IVB cervical cancer across a maximum of 7 weeks of radiation therapy with manually contoured labels including the bladder, cervix, gross tumour volume (GTV), uterus and rectum. The studied approaches accounted for the small dataset and large range of tumour size. Outcomes were evaluated based on the Dice Similarity Coefficients (DSC), the Hausdorff Distance (HD) and the Mean Surface Distance (MSD). The approaches evaluated were shown to have improved segmentation outcomes and required a significantly lower number of model parameters, reducing computer power and memory.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.

Paper Type: both

Primary Subject Area: Segmentation

Secondary Subject Area: Transfer Learning and Domain Adaptation

Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.