FLASH radiotherapy (FLASH-RT) is an emerging radiation therapy technique that delivers ultra-high radiation doses within an extremely short time frame, ensuring effective tumor control while significantly reducing damage to normal tissues. During X-ray FLASH-RT, quantitative reconstruction of the delivered dose distribution in irradiated regions is crucial for dosimetric evaluation and radiotherapy efficacy prediction. X-ray-induced acoustic imaging (XAI), which detects acoustic waves generated by X-ray absorption, has been preliminarily verified in our previous study as an online and noninvasive dosimetry tool during FLASH-RT due to its high sensitivity to radiation deposition. However, conventional reconstruction methods used in XAI (e.g., time reversal) provide a fast solution while dose artifacts remain. To address this challenge, this study proposes a novel physics-informed deep learning framework, termed X-TRUNet, for non-invasive dose reconstruction based on Encoder-Decoder architecture to effectively suppress dose artifacts and improve reconstruction accuracy in X-ray FLASH-RT. Simulation results on standard medical phantoms indicate that the proposed method achieves an RMSE as low as 0.00017, a PSNR exceeding 40 dB, and an SSIM above 0.99 in reconstructed dose images; even under extreme noise conditions (SNR = 0 dB) typical of X-ray FLASH-RT environments, this method still maintains clinically acceptable performance metrics, significantly outperforming conventional reconstruction methods. These studies show the promising potential of the proposed method for achieving high-precision quantitative dosimetry in XAI using an array transducer, potentially enabling in vivo dose verification for X-ray FLASH-RT.
Hu et al. (Sun,) studied this question.