Deep-learning-assisted light scattering spectroscopy for early cancer detection

Light scattering spectroscopy (LSS) provides a non-invasive, in situ detection method without tissue removal, but it can be time-consuming, delaying diagnosis. This study introduces a method combining polarized LSS with a deep learning algorithm. Using a convolutional neural network (CNN) to train a spectral calibration model based on Mie scattering theory, it enables precise particle size distribution calculations. Validated with polystyrene particles, the deep learning algorithm operates five times faster and achieves 50% greater accuracy than the traditional least squares method. Its feasibility was further demonstrated with simulated tissue models of barium sulfate and animal blood mixtures. These advancements in the LSS interrogation algorithm significantly enhance early cancer detection and potentially improve patient prognosis.