Abstract
Early detection of breast microcalcifications, a hallmark of ductal carcinoma in situ, remains a critical clinical challenge, particularly in women with radiographically dense breast tissue. While conventional X-ray mammography presents ionizing radiation risks and reduced sensitivity in dense breasts, standard ultrasonography lacks sufficient acoustic contrast for sub-millimeter mineral deposits. In this study, we present the design, fabrication, and validation of a handheld, miniaturized photoacoustic imaging (PAI) system integrated with a physics-informed deep learning deconvolution network specifically optimized for resolving microcalcifications. The hardware architecture incorporates a compact 128-element piezoelectric transducer array co-aligned with fiber-coupled pulsed laser diode illumination operating at 700 nm and 800 nm. To overcome the acoustic diffraction limits and spatial blurring inherent to miniaturized aperture arrays, we developed a multi-scale convolutional neural network (PA-DeconNet) trained on synthesized and experimentally acquired acoustic point spread functions. In tissue-mimicking phantom experiments embedded with calcium hydroxyapatite particles (30–150 µm), the integrated system achieved an axial resolution of 42 µm and a lateral resolution of 56 µm at depths up to 3.2 cm, representing a 3.2-fold improvement over conventional delay-and-sum reconstruction. In ex vivo human breast biopsy specimens, the platform successfully identified microcalcification clusters with a contrast-to-noise ratio enhancement of 9.8 dB and a sensitivity of 94.1% against micro-computed tomography ground truth. This point-of-care photoacoustic modality offers a non-ionizing, cost-effective adjunct for early breast cancer screening and diagnostic triage.
