Learned feature-based clustering of brain tissue with multimodal optical coherence tomography and photoacoustic elastography

The signals in optical coherence tomography (OCT) and photoacoustic (PA) imaging contain rich information about the intrinsic optical and mechanical properties of tissue, which extends beyond conventional amplitude-based imaging. To extract this information, we propose an unsupervised clustering framework that learns representative features from spatial-domain (i.e., depth domain) OCT profiles and time-domain PA profiles to infer the tissue's underlying properties. Using a modified K-means and spectral clustering approach, we identify latent feature clusters that correspond to distinct tissue types. We validated our method on mouse brain slices, where it clearly delineates major brain regions based on their intrinsic optical and mechanical properties. This method facilitates the differentiation of brain tissue constituents, offering a powerful label-free tool for investigating brain disease pathology.