Neural Network–Based Depth Retrieval of Fluorescent Targets From Wide-Field Fluorescence Intensity Images in Scattering Tissues

Abstract

Fluorescence imaging (FI) is widely used in in vivo and cellular studies. However, accurate depth determination of fluorescent targets in vivo remains challenging due to strong photon scattering and absorption in biological tissues. In this study, we developed and validated a machine learning (ML)-based framework, supported by Monte Carlo (MC) simulations, for accurate depth estimation of near-infrared (NIR) fluorescent targets in turbid media. An optimized MC model was constructed to simulate NIR fluorescence photon propagation, generating datasets of fluorescence images corresponding to target depths between 0.1 and 1 cm. Experimental validation was conducted over a depth range of 0.1–0.7 cm using tissue-mimicking phantoms designed to replicate the optical properties of human skin, as well as on an in vivo model incorporating fluorescent gold nanostructures imaged with computed tomography (CT) and wide-field FI. The results were quantitatively validated using two performance metrics: rounded accuracy and root mean squared error (RMSE).