Deep Learning-Based Computational Adaptive Optics for Photoacoustic Microscopy

Optical-resolution photoacoustic microscopy (OR-PAM) enables high-resolution biomedical imaging but suffers from tissue-induced optical aberrations. The unique photoacoustic detection mechanism in OR-PAM poses challenges for implementing conventional adaptive optics, leaving effective aberration measurement strategies largely underdeveloped. To overcome this limitation, DeepCAO is proposed, a deep learning-based computational adaptive optics framework tailored for OR-PAM. DeepCAO features a two-stage network comprising an untrained denoising module and a supervised end-to-end aberration correction network, jointly trained on experimental and simulated datasets. Validation on simulated and real images—including shallow tissue with known ground truth and deeper tissue with natural aberrations—demonstrates that DeepCAO corrects diverse aberrations. Remarkably, it enables clear visualization of microvessels several hundred micrometers beneath the tissue surface, which would otherwise appear blurred. As a purely computational approach requiring no additional hardware, DeepCAO offers a practical and accessible solution for improving OR-PAM imaging in routine biomedical research.