Abstract #0284

Independent Validation of U-Net Based Breast and Fibroglandular Tissue Segmentation Method on MRI Datasets Acquired Using Different Scanners

Yang Zhang¹, Jeon-Hor Chen^1,2, Kai-Ting Chang¹, Vivian Youngjean Park³, Min Jung Kim³, Siwa Chan⁴, Peter Chang¹, Daniel Chow¹, Alex Luk¹, Tiffany Kwong¹, and Min-Ying Lydia Su¹

¹Department of Radiological Sciences, University of California, Irvine, CA, United States, ²Department of Radiology, E-Da Hospital and I-Shou University, Kaohsiung, Taiwan, ³Department of Radiology and Research Institute of Radiological Science, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea, Democratic People's Republic of, ⁴Department of Medical Imaging, Taichung Tzu-Chi Hospital, Taichung, Taiwan

Segmentation of breast and fibroglandular tissue (FGT) using the U-net architecture was implemented using training MRI from 286 patients, and the developed model was tested in independent validation datasets from 28 healthy women acquired using 4 different MR scanners. The dice similarity coefficient was 0.86 for breast, 0.83 for FGT; and the accuracy was 0.94 for breast and 0.93 for FGT. The results on MRI acquired using different MR scanners were similar. U-net provides a fully automatic, efficient, segmentation method in large MRI datasets for evaluating its role on breast cancer risk assessment and hormonal therapy response prediction.

This abstract and the presentation materials are available to members only; a login is required.

Join Here