BRCA Pathogenic Variants of Breast Cancer Detected by Neural Deep Learning Model

Deep learning is a subset of artificial intelligence (AI) processes focused on using AI neural networks to perform classification and representation learning that takes inspiration from the biologic stacking of neurons to relay a message. Regarding artificial neurons, they are deep layered and trained to process data according to the same architecture as neural networks, thus providing a mesh from which the processing model can identify a given dataset and learn from that identification.

Researchers led by Yi Li and Xiaomin Xiong designed an interpretable deep learning model for detecting BRCA pathogenic variants of breast cancer from hematoxylin- and eosin-stained pathologic images. They aimed to develop a Bi-directional Self-Attention Multiple Instance Learning (BiAMIL) algorithm to detect BRCA status from hematoxylin and eosin pathologic images.

A total of 319 histopathologic slides from 254 patients with breast cancer were included, comprising two dependent cohorts. After image preprocessing, 633,484 tumor tiles from the training dataset were employed to train the self-developed deep learning model. Network performance was evaluated in the internal and external test sets.

BiAMIL achieved area under the curve values of 0.819 (95% CI, 0.673-0.965) in the internal test and 0.817 (95% CI, 0.712-0.923) in the external test set. Class Activation Mapping technique was used to explore the relationship between BRCA gene mutation status, tissue, and cell features.

The tumor-infiltrating lymphocytes and the morphologic characteristics of breast cancer tumor cells appeared to be potential features associated with BRCA status, which is a significant finding and a strong indicator that deep neural network learning can efficiently and accurately predict BRCA status from a dataset. It also provides ample opportunity to further explore this method of disease characterization.