Photoacoustic image synthesis with generative adversarial networks
This work addresses a bottleneck in deep learning-based quantitative PAT by improving synthetic image realism, which is incremental but could enhance training data for medical imaging applications.
The paper tackles the domain gap between real and simulated photoacoustic tomography (PAT) images by proposing a GAN-based approach that generates realistic synthetic images, which were validated to be more realistic than traditional methods in a downstream task.
Photoacoustic tomography (PAT) has the potential to recover morphological and functional tissue properties with high spatial resolution. However, previous attempts to solve the optical inverse problem with supervised machine learning were hampered by the absence of labeled reference data. While this bottleneck has been tackled by simulating training data, the domain gap between real and simulated images remains an unsolved challenge. We propose a novel approach to PAT image synthesis that involves subdividing the challenge of generating plausible simulations into two disjoint problems: (1) Probabilistic generation of realistic tissue morphology, and (2) pixel-wise assignment of corresponding optical and acoustic properties. The former is achieved with Generative Adversarial Networks (GANs) trained on semantically annotated medical imaging data. According to a validation study on a downstream task our approach yields more realistic synthetic images than the traditional model-based approach and could therefore become a fundamental step for deep learning-based quantitative PAT (qPAT).