Focusing light through turbid and thick scattering media via a hybrid genetic neural network algorithm

Over the past decade, techniques for focusing scattered light through dynamic media have made considerable progress, which shows great potential for biomedical imaging, manipulation, and therapy. All of these techniques, however, follow a default principle that the focusing speed must approach the speckle decorrelation time, which hinders applications in living tissues and deep scattering media. Here, we propose a novel perspective that random speckles contain a substantial amount of ballistic light undergoing high-order diffraction over an extended light path. This finding suggests a potential method for focusing light through turbid media, particularly in situations where only a small amount of scattered light can be detected, provided that the ballistic light can be efficiently separated. To achieve this goal, we introduce a hybrid algorithm combining a single-layer neural network for recognizing high-order-diffracted ballistic light with a genetic algorithm for optimizing the roughly focused speckle pattern. The resulting focal spot is demonstrated to remain stable over an extended period, significantly surpassing the limitation imposed by the speckle decorrelation time. Furthermore, we validate the efficacy of our method in ultra-deep (5 cm) turbid liquid environments, where even diffuse light intensity approaches the detection limit of the camera. When the optical thickness of the medium is sufficiently large, both ballistic light and scattered light can be simultaneously converged at the focal spot. This work holds promise for advancing both biomedical imaging and macroscopic active imaging through strongly scattering media.