Author:

Keywords:

Science & Technology, Life Sciences & Biomedicine, Radiology, Nuclear Medicine & Medical Imaging, Breast neoplasms, Magnetic resonance imaging, Neural networks, computer, Image processing, computer-assisted, Image processing, computer-assisted, Neural networks, computer, Humans, Female, Image Processing, Computer-Assisted, Neural Networks, Computer, Breast, Breast Neoplasms, Magnetic Resonance Imaging, 1103 Clinical Sciences, Nuclear Medicine & Medical Imaging, 3202 Clinical sciences

Abstract:

Objectives: To develop a visual ensemble approach using deep convolutional neural networks for 3D segmentation of breast tumors using T1-weighted dynamic contrast-enhanced (T1-DCE) MRI. Methods: Multi-center 3D T1-DCE MRI (n=141) were acquired for a cohort of patients diagnosed with locally advanced or aggressive breast cancer. Tumor lesions of 111 scans were equally divided between two radiologists and segmented for training. The additional 30 scans were segmented independently by both radiologists for testing. Three 3D U-Net models were trained using either post-contrast images or a combination of post-contrast and subtraction images fused either at the image or feature level. Segmentation accuracy was evaluated quantitatively using the Dice Similarity Coefficient (DSC) and the Hausdorff distance (HD95) and scored qualitatively by a radiologist as excellent, useful, or unacceptable. Based on this score, a visual ensemble approach selecting the best segmentation among these three models was proposed. Results: Mean and standard deviation of DSC and HD95 between the two radiologists were equal to 77.8±9.9% and 5.2±5.8 mm. Using the visual ensemble approach, a DSC and HD95 equal to 79.0±15.7% and 14.0±40.1 mm were reached. The qualitative assessment was excellent (resp. excellent or useful) in 53% (resp. 93%). Conclusion: Using subtraction images in addition to post-contrast images provided complementary information for 3D segmentation of breast lesions by convolutional neural networks. A visual ensemble approach allowing the radiologist to select the most optimal segmentation obtained by three 3D U-Net models achieved comparable results to the inter-radiologist agreement, yielding 93% segmented volumes considered excellent or useful.