Background

Quantitative susceptibility mapping (QSM) identifies iron accumulation and myelin loss in smoldering white matter lesions (WMLs). Yet, QSM may be also used to provide a broader understanding of focal and diffuse MS pathology.

Objectives

To study QSM features across WMLs, to assess myelin and axonal loss in WMLs with different QSM features and to quantify QSM pathology in normal-appearing white and cortical grey matter (NAWM, NAGM).

Methods

Ninety-one MS patients (62 RRMS, 29 PMS) and 72 healthy controls (HC) underwent QSM, myelin water imaging (MWI) and multishell diffusion at 3T MRI. In WMLs, cortical lesions (CLs), NAWM and NAGM, we extracted mean QSM, myelin water fraction (MWF) and neurite density index (NDI). WMLs were classified into 5 groups according to their appearance on 3D-EPI QSM: (i) isointense; (ii) with hyperintense rim, Rim+ (iii); with hypointense rim relative to the lesion core, hypo Rim; (iv) hyperintense; (v) hypointense. Mann-Whitney and Kruskal-Wallis test with Dunn’s correction for multiple comparison were used to compare (a) lesion types and (b) specific lesions vs all other WMLs. Voxel-wise comparisons of NAWM QSM were performed using Threshold-Free Cluster Enhancement (TFCE) clustering. Cortical analysis of QSM NAGM and GM-HC was performed using FreeSurfer and compared using a General Linear model (GLM).

Results

Of 1136 WMLs in QSM maps, we detected: (i) 314 (27.6%), (ii) 183 (16.1%), (iii) 16 (1.41%), (iv) 577 (50.8%) and (v) 46 (4.05%) WML. All WML exhibited lower NDI than NAWM and WM-HC (P<0.0001). Isointense lesions exhibited higher NDI (P=0.0115) and MWF (P<0.0001) than other WMLs. Rim + and hyperintense lesions exhibited lower MWF than NAWM and WM-HC (P<0.0001). Rim + lesions showed lower MWF and NDI than other WML types (P<0.001). Hypo Rim+ lesions and hypointense lesions exhibited higher MWF than other WMLs (P=0.0006, P<0.05). Hyperintense lesions exhibited lower MWF than other WMLs types (P<0.01) except Rim+ lesions. TFCE and vertex-wise cortical surface analysis showed areas throughout the NA tissue, where QSM is either lower or higher compared to healthy tissue in HC and in PMS compared to RMS (P<0.01).

Conclusions

QSM is sensitive to diffuse and focal pathology with various myelin and axonal characteristics. We hypothesize that isointense WMLs show high repair activity, hypointense WMLs are remyelinated lesions and hyperintense WMLs are chronic inactive lesions. MRI-histopathology work is ongoing to confirm these findings.

Background

In multiple sclerosis (MS), perilesional chronic inflammation appears on in vivo 3T susceptibility-based magnetic resonance imaging (MRI) as non-gadolinium-enhancing paramagnetic rim lesions (PRL). A higher PRL burden has been recently associated with a more aggressive disease course. The visual detection of PRL by experts is time-consuming and can be subjective.

Objectives

To develop a multimodal convolutional neural network (CNN) capable of automatically detecting PRL on 3D-T2*w-EPI unwrapped phase and 3D-T2w-FLAIR images.

Methods

124 MS cases (87 relapsing remitting MS, 16 primary progressive MS and 21 secondary progressive MS) underwent 3T MRI (MAGNETOM Prisma and MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany). Two neurologists visually inspected FLAIR magnitude and EPI phase images and annotated 462 PRL. 4857 lesions detected by an automatic segmentation (La Rosa et al. 2019) without overlap with PRL were considered non-PRL. The prototype RimNet was built upon two single CNNs, each fed with 3D patches centered on candidate lesions in phase and FLAIR images, respectively. A two-step feature-map fusion, initially after the first convolutional block and then before the fully connected layers, enhances the extraction of low and high-level multimodal features. For comparison, two unimodal CNNs were trained with phase and FLAIR images. The areas under the ROC curve (AUC) were used for evaluation (DeLong et al. 1988). The operating point was set at a lesion-wise specificity of 0.95. The patient-wise assessment was conducted by using a clinically relevant threshold of four rim+ lesions per patient (Absinta et al. 2019).

Results

RimNet (AUC=0.943) outperformed the phase and FLAIR image unimodal networks (AUC=0.913 and 0.855, respectively, P’s <0.0001). At the operating point, RimNet showed higher lesion-wise sensitivity (70.6%) than the unimodal phase network (62.1%), but lower than the experts (77.7%). At the patient level, RimNet performed with sensitivity of 86.8% and specificity of 90.7%. Individual expert ratings yielded averaged sensitivity and specificity values of 76.3% and 99.4%, respectively.

Conclusions

The excellent performance of RimNet supports its further development as an assessment tool to automatically detect PRL in MS. Interestingly, the unimodal FLAIR network performed reasonably well despite the absence of a paramagnetic rim, suggesting that morphometric features such as volume or shape might be a distinguishable feature of PRL.

Background

In multiple sclerosis (MS), perilesional chronic inflammation appears on in vivo 3T susceptibility-based magnetic resonance imaging (MRI) as non-gadolinium-enhancing paramagnetic rim lesions (PRL). A higher PRL burden has been recently associated with a more aggressive disease course. The visual detection of PRL by experts is time-consuming and can be subjective.

Objectives

To develop a multimodal convolutional neural network (CNN) capable of automatically detecting PRL on 3D-T2*w-EPI unwrapped phase and 3D-T2w-FLAIR images.

Methods

124 MS cases (87 relapsing remitting MS, 16 primary progressive MS and 21 secondary progressive MS) underwent 3T MRI (MAGNETOM Prisma and MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany). Two neurologists visually inspected FLAIR magnitude and EPI phase images and annotated 462 PRL. 4857 lesions detected by an automatic segmentation (La Rosa et al. 2019) without overlap with PRL were considered non-PRL. The prototype RimNet was built upon two single CNNs, each fed with 3D patches centered on candidate lesions in phase and FLAIR images, respectively. A two-step feature-map fusion, initially after the first convolutional block and then before the fully connected layers, enhances the extraction of low and high-level multimodal features. For comparison, two unimodal CNNs were trained with phase and FLAIR images. The areas under the ROC curve (AUC) were used for evaluation (DeLong et al. 1988). The operating point was set at a lesion-wise specificity of 0.95. The patient-wise assessment was conducted by using a clinically relevant threshold of four rim+ lesions per patient (Absinta et al. 2019).

Results

RimNet (AUC=0.943) outperformed the phase and FLAIR image unimodal networks (AUC=0.913 and 0.855, respectively, P’s <0.0001). At the operating point, RimNet showed higher lesion-wise sensitivity (70.6%) than the unimodal phase network (62.1%), but lower than the experts (77.7%). At the patient level, RimNet performed with sensitivity of 86.8% and specificity of 90.7%. Individual expert ratings yielded averaged sensitivity and specificity values of 76.3% and 99.4%, respectively.

Conclusions

The excellent performance of RimNet supports its further development as an assessment tool to automatically detect PRL in MS. Interestingly, the unimodal FLAIR network performed reasonably well despite the absence of a paramagnetic rim, suggesting that morphometric features such as volume or shape might be a distinguishable feature of PRL.

Background

Compared to the conventional magnetization-prepared rapid gradient-echo imaging (MPRAGE) MRI, magnetization prepared 2 rapid acquisition gradient echoes (MP2RAGE) (Marques, J. P., 2010) shows a higher brain tissue and lesion contrast in multiple sclerosis (MS) patients. This specialized sequence is, however, mainly limited to research settings and not widely acquired in clinical routine.

Objectives

To synthesize realistic-looking MP2RAGE images from MPRAGE acquisitions via generative adversarial network (GAN) and verify if these improve the performance of automatic MS lesion and brain tissue segmentation tools.

Methods

We propose a GAN inspired by the pix2pix framework (Isola, P., 2018) which takes as input 2D slices of 3D MPRAGE images and generates a synthetic MP2RAGE (synMP2RAGE). Differently from pix2pix, the generator of the GAN combines three loss functions: a pixel-wise L1 loss, an adversarial loss, and a perceptual loss. Our framework is trained on 12 healthy controls and 8 MS patients and tested on 36 MS patients, for which an expert manually delineated cortical and white matter lesions. Imaging was performed with a 3T MRI scanner (Siemens Healthcare, Erlangen, Germany) with a 1x1x1.2 mm resolution. Evaluation is performed with reference-based metrics and through automatic segmentation of MS lesions (La Rosa, F., 2020) and brain tissue (Avants, B.B., 2011).

Results

Considering as reference the acquired MP2RAGE, synMP2RAGE achieves a peak signal-to-noise ratio of 31.39, normalized root mean square error of 0.13, and structural similarity index of 0.98, overperforming the MPRAGE (29.49, 0.17, 0.97, respectively) for all metrics. Compared to the initial MPRAGE it also significantly improves the lesion and tissue segmentation masks in terms of the Dice coefficient and volume difference (p-values < 0.001). On the contrary, no significant differences between the real and synMP2RAGE are found in the patient-wise comparison of the lesions’ segmentation (p-values > 0.05), whereas they are significant between MPRAGE and MP2RAGE (p-value < 0.001).

Conclusions

Our proposed framework successfully translates MPRAGE to MP2RAGE, synthesizing realistic-looking images which improve the performance of automatic segmentation tools tested on MS patients. In accordance with previous claims (Finck, T., 2020), these results confirm that GANs can be helpful in the automatic analysis of MRI images.

Background

Myelin water imaging (MWI) and neurite orientation dispersion and density imaging (NODDI) provide sensitive surrogate markers of myelin and axonal content in lesions and normal-appearing tissue. However, to date, there is scarce information about the relationship of these measures with (i) disability; and (ii) the axonal damage specific biomarker serum neurofilament light chain (sNfL).

Objectives

To explore the correlation of MWI and NODDI measures in MS lesions and in normal-appearing (NA) brain tissue with disability and sNfL.

Methods

Ninety-one MS patients (62 relapsing-remitting MS-RRMS and 29 progressive MS-PMS) underwent MWI and NODDI. Mean myelin water fraction (MWF) and neurite density index (NDI) were extracted in white matter lesions (WMLs), cortical lesions (CLs), NA white matter (NAWM) and cortical NA gray matter (CNAGM). For sNfL, a logarithmic transformation was applied to comply with normality assumption. Correlation studies between MRI measures, sNfL and EDSS were performed using linear models, with age and gender as covariates. The models were performed for the whole sample and for patients with clinical deficits only (EDSS >1).

Results

MWF and NDI did not correlate with EDSS when the entire cohort was considered (P>0.05). However, for those patients with clinical deficits (EDSS> 1), NDI in WMLs was associated with EDSS (NDI: P<0.01, beta=-10.00; N=74). We also found that MWF and NDI in WMLs were related to sNfL (MWF: P<0.01, beta=0.13; NDI: P<0.01, beta=-3.60). Again, this correlation was stronger in patients with EDSS>1 (MWF: P<0.01, beta=0.13; NDI: P <0.01, beta=-3.60).

Conclusions

Imaging surrogate markers of myelin and axon pathology in WML – and not in CLs and NA tissues - are correlated with disability and sNfL. Interestingly, associations between those imaging markers and disability/sNFL were more evident in patients with clinical deficits as compared to those without neurological deficits.

Background

Advanced diffusion-weighted MRI (DW-MRI) sequences, in combination with biophysical models, provide new information on the microstructural properties of the tissue.

Objectives

To investigate the differences in intra-axonal signal fraction (IASF) between perilesional normal-appearing white matter (pl-NAWM), white matter lesions (WML) without (rim-) and with paramagnetic rim (rim+) comparing eight biophysical diffusion models.

Methods

The study included 102 MS patients: RRMS: 66%, SPMS: 18%, PPMS: 16%, mean age 46±14; female 64%, disease duration 12.16±18.18 yrs, median EDSS: 2.5.

DW-MRI data were acquired with 1.8mm isotropic resolution and b-values [0, 700, 1000, 2000, 3000] s/mm².

Lesion masks were generated with a deep-learning-based method and manually corrected if required; pl-NAWM was defined as a region of 3-voxels around each WML; 225 paramagnetic rim lesions were manually identified based on 3D EPI and 2330 were labelled as rim-.

The following microstructural models were applied: Ball and Stick, Ball and Rockets, AMICO-NODDI, SMT-NODDI, MCMDI, NODDIDA, CHARMED, Microstructure Bayesian approach.

Delta (WML - pl-NAWM) was calculated for each WML, and one-side Mann Whitney U was used to compare the delta between models, followed by Bonferroni to correct for multiple testing.

Mean difference and Cohen's d was used to assess differences between lesions with extensive axonal damage (rim+) and other WML (rim-).

Results

All models applied in this study reported low IASF in rim+ WML, medium IASF in rim- WML and relatively high IASF in pl-NAWM. However, a broad spectrum of IASF values was identified from the different models: relatively simple models such as Ball and Stick and CHARMED, showed low delta IASF within lesions, while MCMDI models reported the highest significant difference compared to other models (p<0.0001). The comparison between WML and pl-NAWM mean IASF across models showed that MCDMI exhibited the highest difference (mean 0.13, Cohen’s d 1.34). AMICO-NODDI and SMT-NODDI showed close results (mean difference 0.12/0.12 and Cohen’s d 1.46/1.51).

The models best discriminating IASF between rim+ and rim- lesions were MCMDI and NODIDDA (mean 0.08/0.07, Cohen’s d -0.69/-0.70).

Conclusions

We compared eight WM diffusion models for assessment of intralesional axonal damage in MS patients. The comparison between WML and pl-NAWM showed that robustness of the method, identified with SMT-based and NODDI-based models, it is crucial. For the comparison between lesions with a high level of damage (rim +) and other WML, the diffusivity estimation appeared to play an important role. The method which appeared both robust and able to estimate the diffusivity of the tissue was MCMDI, which performed best in both cases.

Background

Compared to the conventional magnetization-prepared rapid gradient-echo imaging (MPRAGE) MRI, magnetization prepared 2 rapid acquisition gradient echoes (MP2RAGE) (Marques, J. P., 2010) shows a higher brain tissue and lesion contrast in multiple sclerosis (MS) patients. This specialized sequence is, however, mainly limited to research settings and not widely acquired in clinical routine.

Objectives

To synthesize realistic-looking MP2RAGE images from MPRAGE acquisitions via generative adversarial network (GAN) and verify if these improve the performance of automatic MS lesion and brain tissue segmentation tools.

Methods

We propose a GAN inspired by the pix2pix framework (Isola, P., 2018) which takes as input 2D slices of 3D MPRAGE images and generates a synthetic MP2RAGE (synMP2RAGE). Differently from pix2pix, the generator of the GAN combines three loss functions: a pixel-wise L1 loss, an adversarial loss, and a perceptual loss. Our framework is trained on 12 healthy controls and 8 MS patients and tested on 36 MS patients, for which an expert manually delineated cortical and white matter lesions. Imaging was performed with a 3T MRI scanner (Siemens Healthcare, Erlangen, Germany) with a 1x1x1.2 mm resolution. Evaluation is performed with reference-based metrics and through automatic segmentation of MS lesions (La Rosa, F., 2020) and brain tissue (Avants, B.B., 2011).

Results

Considering as reference the acquired MP2RAGE, synMP2RAGE achieves a peak signal-to-noise ratio of 31.39, normalized root mean square error of 0.13, and structural similarity index of 0.98, overperforming the MPRAGE (29.49, 0.17, 0.97, respectively) for all metrics. Compared to the initial MPRAGE it also significantly improves the lesion and tissue segmentation masks in terms of the Dice coefficient and volume difference (p-values < 0.001). On the contrary, no significant differences between the real and synMP2RAGE are found in the patient-wise comparison of the lesions’ segmentation (p-values > 0.05), whereas they are significant between MPRAGE and MP2RAGE (p-value < 0.001).

Conclusions

Our proposed framework successfully translates MPRAGE to MP2RAGE, synthesizing realistic-looking images which improve the performance of automatic segmentation tools tested on MS patients. In accordance with previous claims (Finck, T., 2020), these results confirm that GANs can be helpful in the automatic analysis of MRI images.