Real-Time Phase Retrieval Using On-the-Fly Training of Sample-Specific Surrogate Models

Abstract—Recently, X-ray ptychography is attracting attention as a non-destructive imaging method with high spatial resolution. X-ray ptychography reconstructs a sample image by performing iterative phase retrieval from its diffraction patterns. However, iterative phase retrieval is time-consuming, and thus real-time reconstruction is challenging. One approach to accelerating phase retrieval uses a deep learning based surrogate model to predict the sample. By performing iterative phase retrieval using the model predictions as initial estimates, it is possible to achieve faster reconstruction compared to conventional iterative phase retrieval algorithms. However, although these methods assume to train a surrogate model for individual cases, the training time is often ignored in discussions on performance. Thus, conventional iterative methods might rather be faster if the training time is considered. The objective of this study is to implement real-time phase retrieval using surrogate models, including model training and prediction times. For this purpose, we propose a method to create a sample-specific surrogate model on-the-fly from a subset of observed data and to perform iterative phase retrieval using the predictions of the model. The proposed method is up to 2.8 times faster than conventional iterative phase retrieval, including training dataset creation time and surrogate model training time. In addition, the proposed method also ensures that the reconstructed images are physically correct because it performs the phase retrieval from the surrogate model predictions. This result indicates that it is important to find a good trade-off point between model accuracy and model training time.