AI-assisted diffuse correlation tomography for identifying breast cancer

Abstract Background Microvascular blood flow is more directly relevant to the proliferation of microvascular networks resulting from breast oncogenesis, when compared with the oxy- and deoxy- hemoglobin concentrations in breast tissue. Diffuse correlation tomography (DCT) is an emerging technique for noninvasive measurement of microvascular blood flow, whereas its capability to categorize the benign and malignant breast lesions were not extensively validated thus far, due to the difficulties in instrumentation, image reconstruction algorithms, as well as the appropriate approaches for imaging analyses. Methods In this study, artificial intelligence (AI) assisted DCT instrumentation was constructed based on unique source-detector array and image reconstruction algorithm, with capability for accurately visualizing and diagnosing tumor-induced hemodynamic abnormalities in the microcirculatory network. The DCT blood flow images of human breast were obtained from a total of sixty-one female participants, and AI models were utilized to classify breast lesions. During this process, the DCT blood flow images were either extracted as feature parameters or as the global inputs to the AI models. Results As the validations of DCT instrumentation, the blood flow images obtained from longitudinal monitoring of healthy subject demonstrated stability of DCT measurements on women breasts, with stable bilateral BFI contrast (1.01 ~ 1.12). For the patients with breast diseases, the comprehensive analyses yield an optimal AI-assisted classification for distinguishing between benign and malignant breast lesions, at accuracy rate of 86.4% and an area under the receiver operating characteristic curve of 0.89. Conclusions These outcomes demonstrated the capability of AI-assisted DCT for identifying breast cancer. In contrast to the conventional morphological imaging modalities, the emerging DCT reflects the functional abnormalities that are associated with the cancellous-induced high metabolic demands, thus demonstrating the great potential for early diagnosis and timely therapeutic assessment of breast cancer, e.g., prior to the tumor formation or proliferation of microvascular networks.