BRCANet: A deep hybrid network in predicting BRCA1/2 gene mutation of breast cancer with dynamic contrast-enhanced breast MRI

Autor: Wen-Ming Cao, Xiaojia Wang, Jian Liu, Lili Wang, Xiaodong Zhang, Jiani Pan, Weiwu Ye, Zhanhong Chen, Yabing Zheng, Xiying Shao, Yuanji Xu
Rok vydání: 2022
Předmět:
Zdroj: Journal of Clinical Oncology. 40:e13576-e13576
ISSN: 1527-7755
0732-183X
Popis: e13576 Background: BRCA1/BRCA2, representing as an important genetic biomarker of breast cancer (BCA), can provide clinically significant implications for personalized risk assessment, effective treatment option, and prognostic prediction. Methods: We here proposed the BRCANet, a novel end-to-end convolutional neural network for noninvasively determining BRCA1/BRCA2 mutation by integrating clinical, radiomics and deep learning of dynamic contrast-enhanced (DCE) MRI. BRCANet accepts different forms of medical data including clinicopathologic identifications, high throughput radiomics and deep imaging features of breast MRI using a deep hybrid neural network for data/feature integration. Model training and cross-validation was performed in 132 case-controlled BCA patients from two in two tertiary care hospitals, in which clinicopathologic, genomic and image data of BCA lesions were available and center-standardized for study analysis. Results: Results show that a BRCANet-Plus model, embedded with clinicopathologic, radiomics and deep MRI features achieves an arear under curve of (0.783; 95% confidence intervals [CIs], 0.704 - 0.848) for predicting BRCA1/2 mutation, outperforming the compared state-of-the-art methods, i.e., BRCANet derived from image-only data (0.743; 95% CIs, 0.659 - 0.815; p = 0.037), and BRCARad derived from radiomics-only data (0.734; 95% CIs, 0.649-0.807; p = 0.031). After net benefit evaluation, the proposed BRCANet-Plus shows promise to improve diagnostic performance against conventional clinical or image approaches. Conclusions: Therefore, we concluded the presented deep hybrid approach by integrating multimodal clinical-imaging data, especially breast MRI, have a great potential to predict BRCA1/2 mutational status of BCA. This proof-of-concept strategy can be utilized for studying similar clinical questions.
Databáze: OpenAIRE
Externí odkaz: