1. A novel deep convolutional neural network (DCNN) method was proposed for generating synthetic CT (sCT) from patient MR images.
2. The DCNN model had 27 convolutional layers interleaved with pooling and unpooling layers and 35 million free parameters, which can be trained to learn a direct end-to-end mapping from MR images to their corresponding CTs.
3. The proposed DCNN method produced a mean absolute error (MAE) below 85 HU for 13 of the 18 test subjects, significantly better than the average MAE of 94.5 +/- 17.8 HU for the atlas-based method.
The article is generally reliable and trustworthy in its reporting of the research findings on the use of deep convolutional neural networks (DCNNs) for generating synthetic CT (sCT) from patient MR images. The authors provide detailed information on the methodology used in their study, as well as clear results that demonstrate the efficacy of their approach compared to an atlas-based approach. Furthermore, they provide evidence to support their claims by citing previous studies in the field and providing quantitative results from their experiments.
However, there are some potential biases that should be noted when evaluating this article. For example, it is possible that the authors may have been biased towards their own approach due to its novelty and potential advantages over existing methods; however, they do acknowledge potential limitations such as slow training times and limited data sets used in their experiments. Additionally, while they do mention possible extensions to further improve accuracy or handle multi-sequence MR images, they do not explore these possibilities in detail or provide any evidence for them. Finally, while they note that further validation on dose computation accuracy and on a larger patient cohort is warranted, they do not provide any details on how this could be done or what kind of results would be expected from such an experiment.