Automatic segmentation of tumour and organs at risk in 3D MRI for cervical cancer radiation therapy with anatomical variations

• DOI: 10.1007/s13246-024-01415-y
• Publication Date: 2024-04-24T11:02:14Z
• Authors: Sze-Nung Leung; Shekhar S. Chandra; Karen Lim; Tony Young; Lois Holloway; Jason A. Dowling
• Keywords: Organs at Risk; Male; Radiotherapy Planning, Computer-Assisted; Anatomic Variation; Uterine Cervical Neoplasms; Scientific Paper; Magnetic Resonance Imaging; Automation; 03 medical and health sciences; Imaging, Three-Dimensional; 0302 clinical medicine; Humans; Female

Cervical cancer is a common in women globally, with treatment usually involving radiation therapy (RT). Accurate segmentation for the tumour site and organ-at-risks (OARs) could assist reduction of side effects improve planning efficiency. Magnetic Resonance Imaging (MRI) challenging due to limited amount training data available large inter- intra- patient shape variation OARs. The proposed Masked-Net consists masked encoder within 3D U-Net account dataset, additional dilated layers added performance. A new loss function was introduced consider bounding box during Masked-Net. Transfer learning from male pelvis MRI similar field view included. approaches were compared which widely used image segmentation. consisted 52 volumes obtained 23 patients stage IB IVB cervical across maximum 7 weeks RT manually contoured labels including bladder, cervix, gross volume, uterus rectum. model trained tested 5-fold cross validation. Outcomes evaluated based on Dice Similarity Coefficients (DSC), Hausdorff Distance (HD) Mean Surface (MSD). method accounted small variations OAR sizes an average DSC, HD MSD all anatomical structures 0.790, 30.19mm 3.15mm respectively.