Cross-site Validation of AI Segmentation and Harmonization in Breast MRI
Authors:
- 25 September 2024
- journal article
- research article
- Published by Springer Nature in Journal of Imaging Informatics in Medicine
- Vol. 38 (3) , 1642-1652
Abstract
This work aims to perform a cross-site validation of automated segmentation for breast cancers in MRI and to compare the performance to radiologists. A three-dimensional (3D) U-Net was trained to segment cancers in dynamic contrast-enhanced axial MRIs using a large dataset from Site 1 (n = 15,266; 449 malignant and 14,817 benign). Performance was validated on site-specific test data from this and two additional sites, and common publicly available testing data. Four radiologists from each of the three clinical sites provided two-dimensional (2D) segmentations as ground truth. Segmentation performance did not differ between the network and radiologists on the test data from Sites 1 and 2 or the common public data (median Dice score Site 1, network 0.86 vs. radiologist 0.85, n = 114; Site 2, 0.91 vs. 0.91, n = 50; common: 0.93 vs. 0.90). For Site 3, an affine input layer was fine-tuned using segmentation labels, resulting in comparable performance between the network and radiologist (0.88 vs. 0.89, n = 42). Radiologist performance differed on the common test data, and the network numerically outperformed 11 of the 12 radiologists (median Dice: 0.85–0.94, n = 20). In conclusion, a deep network with a novel supervised harmonization technique matches radiologists’ performance in MRI tumor segmentation across clinical sites. We make code and weights publicly available to promote reproducible AI in radiology.Keywords
Funding Information
- Foundation for the National Institutes of Health (NIH R01CA247910, NIH P30CA008748, NIH R01EB022720, NIH R01MH111896)
- National Science Foundation
- Breast Cancer Research Foundation
This publication has 20 references indexed in Scilit:
- Annotation-efficient deep learning for automatic medical image segmentationNature Communications, 2021
- Convolutional neural networks for breast cancer detection in mammography: A surveyComputers in Biology and Medicine, 2021
- Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approachNature Medicine, 2021
- International evaluation of an AI system for breast cancer screeningNature, 2020
- Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer ScreeningIEEE Transactions on Medical Imaging, 2019
- A machine learning approach to radiogenomics of breast cancer: a study of 922 subjects and 529 DCE-MRI featuresBritish Journal of Cancer, 2018
- Deep learning for segmentation of brain tumors: Impact of cross‐institutional training and testingMedical Physics, 2018
- Comparative Studies of Different Gadolinium Agents in Brain Tumors: Differences between Gadolinium Chelates and Their Possible Influence on Imaging FeaturesAmerican Journal of Neuroradiology, 2010
- Extracellular gadolinium-based contrast media: An overviewEuropean Journal of Radiology, 2008
- Nonrigid registration using free-form deformations: application to breast MR imagesIEEE Transactions on Medical Imaging, 1999