Recently, convolutional neural networks (CNNs) have been proposed as a method for deformable image registration,offering a variety of potential advantages compared to physical model-based methods, including faster runtime and abilityto learn complicated functions without explicit models. A persistent question for CNNs is the uncertainty in their behaviorwhen the image statistics (e.g., noise and resolution) of the test data deviate from those of the training data. In this workwe investigated the influence of statistical properties of image noise (in CT, for example, related to radiation dose). Wetrained registration networks over a range of dose levels and evaluated registration performance (target registration error,TRE) as the statistics of the test data deviated from that of the training data. Generally, registration performance wasoptimal when the statistics of the test data matched that of the training data. Furthermore, TRE was found to be limited bythe highest dose training data, with no improvement in TRE for test images of higher dose than that in the training data.Understanding and quantifying the relationship between statistical aspects of the training and test data – and the failuremodes caused by statistical mismatch – is an important step in the development of CNN-based registration methods. Thiswork provided new insight on the optima and tradeoffs with respect to image noise (dose), providing important guidancein building training sets that are best-suited to particular imaging conditions and applications.
展开▼
