Integrating machine learning with panoramic photoacoustic computed tomography for improved breast lesion analysis (Conference Presentation)

Screening of breast cancer is crucial due to its high prevalence and mortality. Traditional methods like mammography, ultrasonography, and MRI have limitations, including radiation exposure, operator dependency, and high costs. Photoacoustic computed tomography (PACT) emerges as a promising alternative, combining light and ultrasound to detect tumor-related vascular changes with high resolution. This study presents a comprehensive workflow utilizing panoramic PACT to detect and diagnose breast lesions, providing detailed angiograms independent of breast density and supporting longitudinal monitoring. By analyzing 78 breasts from 39 patients, we developed machine learning classifiers that distinguish healthy from suspicious tissues, achieving a maximum area under the receiver operating characteristic curve (AUROC) of 0.89, comparable or superior to conventional methods. Additionally, we identified 13 features to differentiate malignant from benign lesions, reducing unnecessary biopsies. Our study highlights PACT as a non-invasive, sensitive, and cost-effective tool for breast lesion screening and diagnosis.