Uncertainty-Aware and Lesion-Specific Image Synthesis in Multiple Sclerosis Magnetic Resonance Imaging: A Multicentric Validation Study

Generative adversarial networks (GANs) can synthesize high-contrast MRI from lower-contrast input. Targeted translation of parenchymal lesions in multiple sclerosis (MS), as well as visualization of model confidence further augment their utility, provided that the GAN generalizes reliably across different scanners. We here investigate the generalizability of a refined GAN for synthesizing high-contrast double inversion recovery (DIR) images and propose the use of uncertainty maps to further enhance its clinical utility and trustworthiness. A GAN was trained to synthesize DIR from input fluid-attenuated inversion recovery (FLAIR) and T1w of 50 MS patients (training data). In another 50 patients (test data), two blinded readers (R1 and R2) independently quantified lesions in synthetic DIR (synthDIR), acquired DIR (trueDIR) and FLAIR. Of the 50 test patients, 20 were acquired on the same scanner as training data (internal data), while 30 were scanned at different scanners with heterogeneous field strengths and protocols (external data). Lesion-to-Background ratios (LBR) for MS-lesions vs. normal appearing white matter, as well as image quality parameters were calculated. Uncertainty maps were generated to visualize model confidence. Significantly more MS-specific lesions were found in synthDIR compared to FLAIR (R1: 26.7 ± 2.6 vs. 22.5 ± 2.2 p < 0.0001; R2: 22.8 ± 2.2 vs. 19.9 ± 2.0, p = 0.0005). While trueDIR remained superior to synthDIR in R1 [28.6 ± 2.9 vs. 26.7 ± 2.6 (p = 0.0021)], both sequences showed comparable lesion conspicuity in R2 [23.3 ± 2.4 vs. 22.8 ± 2.2 (p = 0.98)]. Importantly, improvements in lesion counts were similar in internal and external data. Measurements of LBR confirmed that lesion-focused GAN training significantly improved lesion conspicuity. The use of uncertainty maps furthermore helped discriminate between MS lesions and artifacts. In conclusion, this multicentric study confirms the external validity of a lesion-focused Deep-Learning tool aimed at MS imaging. When implemented, uncertainty maps are promising to increase the trustworthiness of synthetic MRI.

Detection of cortical malformations using enhanced synthetic contrast images derived from quantitative T1 maps

The detection of cortical malformations in conventional MR images can be challenging. Prominent examples are focal cortical dysplasias (FCD), the most common cause of drug-resistant focal epilepsy. The two main MRI hallmarks of cortical malformations are increased cortical thickness and blurring of the gray (GM) and white matter (WM) junction. The purpose of this study was to derive synthetic anatomies from quantitative T1 maps for the improved display of the above imaging characteristics in individual patients. On the basis of a T1 map, a mask comprising pixels with T1 values characteristic for GM is created from which the local cortical extent (CE) is determined. The local smoothness (SM) of the GM-WM junctions is derived from the T1 gradient. For display of cortical malformations, the resulting CE and SM maps serve to enhance local intensities in synthetic double inversion recovery (DIR) images calculated from the T1 map. The resulting CE- and/or SM-enhanced DIR images appear hyperintense at the site of cortical malformations, thus facilitating FCD detection in epilepsy patients. However, false positives may arise in areas with naturally elevated CE and/or SM, such as large GM structures and perivascular spaces. In summary, the proposed method facilitates the detection of cortical abnormalities such as cortical thickening and blurring of the GM-WM junction which are typical FCD markers. Still, subject motion artifacts, perivascular spaces, and large normal GM structures may also yield signal hyperintensity in the enhanced synthetic DIR images, requiring careful comparison with clinical MR images by an experienced neuroradiologist to exclude false positives.

A case of epilepsy in multiple sclerosis: Three-dimensional double inversion recovery sequences revealed cortical dysplasia

In epileptic patients with multiple sclerosis (MS), cortical lesions have been suggested to cause seizures. In brain magnetic resonance imaging (MRI), double inversion recovery (DIR) sequences are generally used to evaluate MS cortical disease burden. We present the case of a woman, diagnosed with MS, suffering from drug-resistant partial seizures initially attributed to MS. The patient underwent many MRI exams, but only by means of high-resolution three-dimensional DIR sequences was a focal cortical dysplasia discovered. The MRI findings and FDG-PET/CT supported the diagnosis. This case recommends the use of DIR sequences both in patients with suspect epileptogenic lesions not detected with routine MRI protocols and in epileptic patient with MS, before ascribing seizures to MS.

[Relationship between Blurring and Refocus Flip Angle in 3D-double Inversion Recovery MRI]

It is important to optimize imaging parameters in 3D-double inversion recovery (DIR) magnetic resonance imaging (MRI) for detecting cortical micro lesions. However, inadequate parameters markedly raise blurring in 3DDIR MRI. The purpose of this study was to evaluate the relationship between the blurring and refocus flip angle (RFA) in 3D-DIR MRI. White matter attenuated inversion recovery (WAIR) images as a test sample were obtained by 1.5T MRI with various RFA settings (30°, 40°, 60°, 100°, 140°, 180°, and variable refocus flip angle (VRFA)). Optimal RFA was evaluated using Scheffé's method (Nakaya changing method) by five observers. The results of average preferences indicated that RFA settings of under the 60° of RFA or VRFA suppressed the blurring in 3DDIR MRI. The yard sticks of RFAs of 30° and 40° were significantly higher than the yard sticks of other RFAs (p<0.01). For detecting cortical microlesions, it is very important to obtain WAIR images with no blurring. Using low RFA or VRFA didn't cause significant differences of signal intensity between high-frequency region and low-frequency region in k-space of 3D-DIR MRI. Therefore, it is recommended to set lower RFA (under 60° or VRFA) for suppressing blur in 3D-DIR MRI.