Neural network-based tissue attenuation correction in dual-band cherenkov imaging for dose distribution verification.

PURPOSE: This study aims to address the attenuation of Cherenkov signals caused by tissue heterogeneity in Cherenkov imaging and to improve the accuracy of dose quantification by applying an advanced attenuation correction method to the Cherenkov light. METHODS: The Neural Network-Based Attenuation Correction Model With Dual-Band Cherenkov Images was proposed. Dual-band Cherenkov images with wavelength ranges of 550 ± 50 nm and 660 ± 40 nm were acquired using an EMCCD camera, and optical feature data were extracted as input for the network. Attenuation correction factors for each tissue region were predicted and applied to the original images to generate more accurate dose distributions. Experiments were conducted under both uniform and non-uniform dose conditions, and the performance of the model was evaluated by comparing dose distribution uniformity and its correlation with planned dose levels before and after correction. RESULTS: Under uniform dose fields, the uniformity of dose distribution derived from Cherenkov signals improved markedly after correction, with the coefficient of variation reduced from 20.8%-33.8% before correction to 1.2%-1.8% after correction. In non-uniform dose fields, this attenuation correction method significantly reduces tissue-induced Cherenkov light attenuation, and the dose distribution mapped by the corrected Cherenkov light is highly consistent with the actual surface dose. CONCLUSIONS: The proposed method can effectively compensate for Cherenkov signal attenuation caused by tissue heterogeneity, thereby improving the accuracy of dose quantification and laying a foundation for its future application in precise quantitative dose assessment.