A self-supervised learning approach for high-resolution diffuse optical tomography using neural fields

Abstract

Diffuse optical tomography (DOT) has shown promise in biomedical research, such as breast cancer diagnostics and brain imaging, by reconstructing hidden objects within scattering media. However, the conventional reconstruction framework faces challenges due to the highly ill-posed inverse problem of reconstructing optical properties. This work introduces a novel approach, neural field-based diffuse optical tomography (NeuDOT), which leverages a multi-layer perceptron (MLP) to learn an implicit function that maps spatial coordinates to their corresponding optical absorption coefficients. The performance of the NeuDOT method has been evaluated through several phantom studies, demonstrating its potential for high spatial resolution DOT reconstruction