BACKGROUND AND PURPOSE: Synthetic FLAIR images are of lower quality than conventional FLAIR images. Here, we aimed to improve the synthetic FLAIR image quality using deep learning with pixel-by-pixel translation through conditional generative adversarial network training. MATERIALS AND METHODS: Forty patients with MS were prospectively included and scanned (3T) to acquire synthetic MR imaging and conventional FLAIR images. Synthetic FLAIR images were created with the SyMRI software. Acquired data were divided into 30 training and 10 test datasets. A conditional generative adversarial network was trained to generate improved FLAIR images from raw synthetic MR imaging data using conventional FLAIR images as targets. The peak signal-to-noise ratio, normalized root mean square error, and the Dice index of MS lesion maps were calculated for synthetic and deep learning FLAIR images against conventional FLAIR images, respectively. Lesion conspicuity and the existence of artifacts were visually assessed. RESULTS: The peak signal-to-noise ratio and normalized root mean square error were significantly higher and lower, respectively, in generated-versus-synthetic FLAIR images in aggregate intracranial tissues and all tissue segments (all P < .001). The Dice index of lesion maps and visual lesion conspicuity were comparable between generated and synthetic FLAIR images (P = 1 and .59, respectively). Generated FLAIR images showed fewer granular artifacts (P = .003) and swelling artifacts (in all cases) than synthetic FLAIR images. CONCLUSIONS: Using deep learning, we improved the synthetic FLAIR image quality by generating FLAIR images that have contrast closer to that of conventional FLAIR images and fewer granular and swelling artifacts, while preserving the lesion contrast.

Osteoma of the internal auditory canal (IAC) is an uncommon benign bone tumor. Its imaging features may be similar to other IAC lesions, such as vestibular schwannomas that are benign and usually slow-growing but sometimes life-threatening tumors. Thus, detecting IAC lesions and differentiating osteoma from other IAC lesions are both important clinically. We report a case of misdiagnosis of an IAC osteoma as an IAC schwannoma based on magnetic resonance (MR) imaging using the three-dimensional constructive interference in steady state (CISS) sequence instead of T1-weighted MR imaging with gadolinium. We also review 17 cases of IAC osteomas reported in the past 22 years. A 61-year-old female was admitted to our department with IAC lesion incidentally discovered by the CISS sequence. The lesion was diagnosed as an IAC schwannoma, and was followed up annually under “wait and scan” management. Follow-up T1-weighted MR imaging with gadolinium showed no enhancement of the tumor, and additional computed tomography (CT) of the temporal bone showed a solitary pedunculated bony lesion, resulting in the diagnosis of IAC osteoma. The CISS sequence is useful for detecting small IAC lesions, such as vestibular schwannomas. However, the CISS sequence has limitations for qualitative diagnosis and can misdiagnose osteomas as schwannomas. Use of the CISS sequence without T1-weighted MR imaging with gadolinium for the screening of a lesion of the IAC and cerebellopontine angle should consider the possibility of IAC osteomas, and temporal bone CT or T1-weighted MR imaging with gadolinium should be performed when an IAC lesion is detected.

Purpose: A general problem of machine-learning algorithms based on the convolutional neural network (CNN) technique is that the reason for the output judgement is unclear. The purpose of this study was to introduce a strategy that may facilitate better understanding of how and why a specific judgement was made by the algorithm. The strategy is to preprocess the input image data in different ways to highlight the most important aspects of the images for reaching the output judgement.Materials and Methods: T2-weighted brain image series falling into two age-ranges were used. Classifying each series into one of the two age-ranges was the given task for the CNN model. The images from each series were preprocessed in five different ways to generate five different image sets: (1) subimages from the inner area of the brain, (2) subimages from the periphery of the brain, (3–5) subimages of brain parenchyma, gray matter area, and white matter area, respectively, extracted from the subimages of (2). The CNN model was trained and tested in five different ways using one of these image sets. The network architecture and all the parameters for training and testing remained unchanged.Results: The judgement accuracy achieved by training was different when the image set used for training was different. Some of the differences was statistically significant. The judgement accuracy decreased significantly when either extra-parenchymal or gray matter area was removed from the periphery of the brain (P < 0.05).Conclusion: The proposed strategy may help visualize what features of the images were important for the algorithm to reach correct judgement, helping humans to understand how and why a particular judgement was made by a CNN.

Purpose: The purpose of this study was to evaluate the suitability of two phantoms, one made of capillary plates and the other polyethylene fibers, for assessing the quality of diffusion tensor imaging (DTI). Methods: The first phantom was a stack of glass capillary plates with many parallel micropores (CP). The second phantom was a bundle of polyethylene fiber Dyneema held together with a thermal shrinkage tube (Dy). High resolution multi-shot echo planar imaging (EPI) DTI acquisitions were performed at b-values of 0 and 1000 s/mm(2) and diffusion-times (T-diff) of 37.7 and 97.7 ms on a preclinical 7T MRI scanner. Thirty diffusion-encoding directions were used. The data were used to calculate the fractional anisotropy (FA), mean diffusivity (MD), and angular dispersion (AD). Further acquisitions were performed at b-values from 0 to 8000 s/mm(2) in 14 steps with the diffusion gradient applied parallel (axial) and perpendicular (radial) to the Z direction. On the other hand, the data acquired with a 3T MRI scanner were used to confirm that measurements on a clinical machine are consistent with the 7T MRI results. Results: The dependence of FA, MD and AD on Tiff was smaller for the Dy than for the CPs. The b-value-dependent signal attenuations in the axial direction at T-diff = 37.7 and 97.7 ms were similar for both phantoms. In the radial direction, Dy demonstrated similar b-value attenuation to that of in vivo tissue for both T(diff)s, but the attenuation for the CPs was affected by the change in T-diff. Parameter estimates were similar for 3T and 7T MRI. Conclusion: The characteristics of the CP indicate that it can be used as a restricted-diffusion dominant phantom, while the characteristics of the Dy suggest that it can be used as a hindered-diffusion dominant phantom. Dy may be more suitable than CP for DTI quality control.