The purpose of this retrospective study was to develop and evaluate the performance of U-Net to determine whether U-Net-based deep learning could accurately perform fully automated localization and segmentation of cervical tumors in MR images, as well as the robustness of extracting apparent diffusion coefficient (ADC) radiomics features.
- U-Net-based deep learning can perform accurate fully automated localization and segmentation of cervical cancer in diffusion-weighted MR images.
- Combining b0, b1000, and apparent diffusion coefficient (ADC) images exhibited the highest accuracy in fully automated localization.
- First-order radiomics feature extraction from whole tumor volume was robust and could thus potentially be used for longitudinal monitoring of treatment responses.
Authors: Yu-Chun Lin, Chia-Hung Lin, Hsin-Ying Lu, Hsin-Ju Chiang, Ho-Kai Wang, Yu-Ting Huang, Shu-Hang Ng, Ji-Hong Hong, Tzu-Chen Yen, Chyong-Huey Lai & Gigin Lin