Deep Fluorescence Imaging by Laser‐Scanning Excitation and Artificial Neural Network Processing

Abstract Fluorescence imaging using charge‐coupled device cameras has become the prevailing method for the investigation of cancer drug efficacy in in vivo preclinical trials involving animal models. The lack of imaging depth, however, limits the use of fluorescence techniques to small murine models, and compels large animal models to rely on more‐costly magnetic resonance imaging or positron emission tomography techniques. Here, a wide field‐of‐view fluorescence imaging technique that uses near‐infrared (NIR) laser‐scanning excitation and efficient quantum dot photoluminescence is developed to achieve deep imaging across thick animal tissues. The smaller excitation volume from the scanning laser beam minimizes undesired background fluorescence, and allows signals to be detected and resolved at large tissue depths exceeding 10 mm. By implementing an artificial neural network algorithm, a twofold enhancement in imaging resolution is further demonstrated, which paves a promising way forward for the use of machine intelligence to enhance imaging quality in highly scattering media. The superior contrast of the imaging method is further corroborated by the imaging of a human palm, where bone structures are, for the first time, distinguishable using a NIR technique. In combination, the laser‐scanning approach and the artificial neural network image‐processing can constitute a low‐cost, yet powerful methodology for performing noninvasive deep‐imaging in larger animal models or human tissues.