TIME: Tractography-Informed myelin estimation

Highlights

• •Myelin loss within WM lesions is less explored than NAWM or global WM estimates.
• •TIME quantifies tract-specific myelin loss, combining tractography with myelin maps.
• •Myelin loss is referenced to healthy tissue within the same white matter bundle.
• •TIME was applied to MS datasets to assess baseline and longitudinal myelin loss.
• •TIME allows enhanced sensitivity to clinically relevant myelin loss in MS lesions.

Abstract

Investigating myelin integrity within multiple sclerosis (MS) lesions and in normal-appearing white matter is crucial for understanding demyelination and remyelination processes. While most approaches assess global myelin changes or compare lesions with homologous regions in healthy controls, they do not allow direct within-tract comparisons between lesional and non-lesional tissue.

We introduce the tractography-informed myelin estimate (TIME), a novel map designed to quantify tract-specific myelin loss. TIME integrates tractography with myelin-sensitive imaging, such as myelin volume fraction, to compare lesional and non-lesional segments within the same white matter tract. By modeling local deviations from the expected myelin volume fraction signal along streamlines, TIME captures tract-specific myelin damage while accounting for within-tract variability. TIME is based on a microstructure-informed tractography framework, with an extra compartment to model signal loss caused by lesions.

We evaluated TIME in 159 MS patients, assessing its association with neurological disability at baseline and longitudinally over a median follow-up of two years. At baseline, higher myelin loss captured by TIME was significantly associated with worse disability (β = 0.14, p = 0.015). Longitudinally, greater baseline disability predicted faster TIME-quantified myelin loss, which was in turn associated with a higher risk of disability worsening. In contrast, lesion-averaged myelin volume fraction showed no significant associations with either baseline disability or its progression.

TIME provides a detailed, tract-specific assessment of myelin damage, providing greater sensitivity than conventional metrics, highlighting its potential as a biomarker in MS.

1. Introduction

Myelin is a lipid-rich substance that forms sheaths around axons, protecting nerve fibers within the central nervous system and ensuring the efficient transmission of electrical impulses (Morell & Quarles, 1999). In multiple sclerosis (MS), the myelin sheath is targeted by inflammatory processes, leading to the formation of demyelinating lesions or plaques. These lesions disrupt the normal transmission of electrical impulses, leading to a wide range of neurological symptoms characteristic of MS (Ghasemi et al., 2017, Lubetzki and Stankoff, 2014, Reich et al., 2018). In MS, demyelination triggers a response from the central nervous system aimed at repairing the damaged myelin. Initially, remyelination can be effective in the early stages of MS. However, as the disease progresses, remyelination typically becomes less efficient. Chronic demyelination eventually leads to irreversible axonal damage, which is considered the primary contributor to the accumulation of disability in MS (Ghasemi et al., 2017, Lassmann, 2022, Podbielska et al., 2013, Reich et al., 2018). Ongoing MS research aims to explore strategies to repair damaged myelin, reduce inflammation, and manage symptoms. Understanding the properties and dynamics of myelin is central to this effort, offering critical insights into the processes of demyelination and remyelination, and potentially informing the development of more effective therapies.

Several studies have investigated global myelin alterations in MS using multiple advanced neuroimaging techniques, examining associations between global myelin values and clinical disability scales (Bezukladova et al., 2020, Bonnier et al., 2019, Cagol et al., 2024, Gracien et al., 2016, Kolind et al., 2012, Margoni et al., 2022, Oreja-Guevara et al., 2006, Vrenken et al., 2006, York et al., 2022).

Another research, performed by Rahmanzadeh et al. (2022), looked at remyelinated lesions using a technique called Quantitative Susceptibility Mapping (QSM). By using this technique, they were able to identify five types of lesions: iso-intense, hypo-intense, hyperintense, lesions with hypo-intense rims, and lesions with paramagnetic rim lesions (Granziera et al., 2020, Rahmanzadeh et al., 2022). The study found that hypo- and iso-intense lesions had higher mean myelin water fraction and neurite density index values than other QSM lesion types. Additionally, after a 2-year follow-up, these hypo-/iso-intense lesions showed increased myelin water fraction (Rahmanzadeh et al., 2022).

Although these studies have advanced our understanding of global myelin alterations and remyelinated lesion subtypes, few have directly investigated the clinical relevance of myelin damage within white matter lesions. Most of the literature has focused on normal-appearing white matter or on global myelin metrics (Bezukladova et al., 2020, Bonnier et al., 2019, Cagol et al., 2024, Margoni et al., 2022, Oreja-Guevara et al., 2006, Vrenken et al., 2006, York et al., 2022). Some studies have reported focal myelin differences by comparing lesional areas in patients with homologous regions in healthy controls, suggesting that localized myelin loss is detectable (Bonnier et al., 2019, Chen et al., 2023, Kolind et al., 2012, Margoni et al., 2022). However, such approaches do not enable direct comparison between lesional and non-lesional tissue within the same subject and along the same white matter pathway.

Building upon these insights, we propose a novel tractography-informed map sensitive to myelin damage inside the lesions. Similar to previous approaches (Bonnier et al., 2019, Chen et al., 2023), this map quantifies focal myelin loss; however, it does so relative to the myelin content of the same bundle in unaffected brain regions. Specifically, the map we propose is the output of an extension of Myelin Streamlines Decomposition (MySD) framework (Bosticardo et al., 2023, Schiavi et al., 2022). MySD is part of the microstructure-informed tractography techniques. These techniques assume that the fibers reconstructed with tractography, also called streamlines, represent groups of axons following a common trajectory while maintaining constant microstructural properties along their path. In the presence of focal pathology such as MS lesions, however, this assumption no longer holds. To address this, the model was extended by introducing a lesion-specific compartment through a multi-compartment formulation, which enables the estimation of non-constant microstructural properties along the affected tracts (for more information, we refer to (Bosticardo et al., 2025)).

In this study, we used Magnetization Transfer saturation (MTsat)-derived Myelin Volume Fraction (MVF) maps as input to the multi-compartment MySD model to quantify myelin differences along the same streamline evaluated in normal-appearing white matter voxels and in voxels affected by focal lesions. Given the additive nature of the MVF signal, the framework infers the expected signal contribution of each streamline segment. Deviations between observed and expected MVF values in lesional voxels are interpreted as evidence of focal myelin loss. We refer to this measure of focal myelin loss as the Tractography-Informed Myelin Estimate (TIME). To validate the clinical relevance of TIME, we computed, for each subject, the average TIME value across all the white matter lesions (WMLs) and assessed its association with baseline Expanded Disability Status Scale (EDSS). For comparison, we repeated the analysis using the average MVF within WMLs, a standard lesion-based myelin mapping approach commonly adopted in the literature. At baseline, we also computed the ratio of values assessed using TIME or the standard method within QSM-defined remyelinated lesions relative to all WMLs and tested its association with EDSS. Finally, we assessed whether baseline EDSS predicted longitudinal changes in TIME and the standard method, and whether these changes were associated with clinical progression.

2. Methods

The multi-compartment MySD model produces two output maps: one representing the bundle-specific myelin content, and another capturing the lesion-specific deviation from the expected signal, which we define as the TIME.

To generate these maps, each streamline is segmented into voxel-wise fragments, and the measured MVF signal is modeled as a linear combination of contributions from a myelin compartment and, where applicable, a lesion compartment. In healthy tissue, myelin content is assumed to be constant along the streamline. In contrast, in voxels intersected by streamlines and marked as lesional, the lesion compartment models the signal component that cannot be explained by normal myelin content alone, thereby quantifying the extent of damage.

This approach allows for the estimation of myelin content along white matter tracts across both normal and pathological tissues, enabling the identification of focal myelin loss within lesions in a bundle-specific manner.

Fig. 1 illustrates how TIME values are derived from the interaction between streamlines and MVF measurements.

Fig. 1.

Graphic illustration of the Tractography-Informed Myelin Estimate (TIME) computation. A streamline passing through a lesion is internally divided into voxel-wise fragments. The myelin content along a streamline is assumed to be constant across all voxels it traverses (in this example, equal to ten). In voxels where the signal deviates from this expected value (e.g., the light-blue voxel showing a value of six instead of ten), the difference is attributed to the lesion compartment. This discrepancy is reflected in the TIME map, where the same light-blue voxel is assigned a value of four, representing the estimated myelin loss in that voxel relative to its neighbors along the streamline. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.1. Dataset

We included data from 159 patients with MS (96 (60 %) females, mean (SD) age 46.27 (14.33) years, median [interquartile range] EDSS 3.0 [1.5–4.75], 94 (59 %) with relapsing remitting (RR) MS and 65 (41 %) with progressive MS). Full demographic and clinical characteristics are reported in Table 1. All participants underwent brain MRI scans at baseline and after 2 years (+/- 3 months) using a standardized MRI protocol on a 3 T MRI whole-body system (Magnetom Prisma; Siemens Healthineers, Erlangen, Germany) with a 64-channel head and neck coil. The acquisition protocol included: (1) 3D FLAIR (repetition time (TR) / echo time (TE) / inversion time (TI) = 5000/386/1800  ms) and MP2RAGE (TR/TI1/TI2 = 5000/700/2500 ms), both with (1.0 mm)³ isotropic spatial resolution; (2) echo planar imaging (EPI) based multi-shell diffusion weighted imaging (multiband = 2, TR/TE = 4500/75 ms, diffusion sensitization with pulse duration/distance δ/Δ = 19/36 ms; (1.8 mm)³ isotropic spatial resolution with b values 700/1000/2000/3000 s/mm² and 6/20/45/66 diffusion directions, respectively, per shell, and 12 measurements at b value 0 s/mm² with both anterior to posterior and reversed phase encoding; (3) 3D segmented EPI with submillimeter isotropic resolution (0.67 mm)³ (TR/TE = 64 ms/35 ms), providing both T2* magnitude and phase contrast; and (4) three variants of a 3D FLASH (RF spoiled GRE) sequence with (1.33  mm)³ isotropic resolution, PPF = 6/8; SPF = 6/8, GRAPPA = 2: T1-weighted (T1w) (TR/TE = 11/4.92  ms, alpha = 15°), Proton Density weighted (TR/TE = 25/4.92  ms, alpha = 5°), MT-weighted [TR/TE = 25/4.92  ms, alpha = 5°, Gaussian MT pulse Delta_f = 2.2KHz as in (Helms et al., 2008a, Helms et al., 2008b)]. B1 maps were acquired to correct the effects of radiofrequency transmit inhomogeneities on the quantitative maps, employing the steady state free precession-based B1-TRAP approach (Ganter et al., 2013).

Table 1.

Demographic and clinical characteristics of the study cohort. Values are expressed as mean (standard deviation, SD), median [interquartile range, IQR], or number (percentage, %) as appropriate. EDSS: Expanded Disability Status Scale; PPMS: primary progressive multiple sclerosis; RRMS: relapsing–remitting multiple sclerosis; SPMS: secondary progressive multiple sclerosis; DMT: disease-modifying therapy.

	Level	Overall
n		159
Age (mean (SD)), years		46.27 (14.33)
Disease Duration (median [IQR]), years		6.15 [0.93, 16.97]
EDSS (median [IQR])		3.00 [1.50, 4.75]
Sex	Female	96 (60.4)
	Male	63 (39.6)
Diagnosis (%)	PPMS	22 (13.8)
	RRMS	94 (59.1)
	SPMS	43 (27.0)
DMT (%)	Dimethyl fumarate	15 (9.4)
	Fingolimod	18 (11.3)
	Glatiramer acetate	1 (0.6)
	Interferon beta preparations	6 (3.8)
	Natalizumab	5 (3.1)
	Ocrelizumab	68 (42.8)
	Rituximab	16 (10.1)
	Siponimod	2 (1.3)
	Teriflunomide	4 (2.5)
	Untreated	24 (15.1)

Inclusion criteria for MS patients were: (1) age between 18 and 75 years; (2) diagnosis of MS according to the McDonald criteria (Thompson et al., 2018); (3) absence of neurological or psychiatric comorbidities; and (4) absence of contraindications to MRI. The study was approved by the Ethics Committee (IRB of Northwest Switzerland), and all participants entered the study after providing written consent.

2.2. MRI analysis

Diffusion MRI (dMRI) data were pre-processed to reduce artifacts from noise (Veraart et al., 2016a, Veraart et al., 2016b), eddy currents (Andersson & Sotiropoulos, 2016), motion, and EPI distortions (Andersson et al., 2003; S. M. Smith et al., 2004) using MRtrix3 (Tournier et al., 2019) and FSL (Jenkinson et al., 2012, Woolrich et al., 2009). dMRI images were upsampled to match the 1x1x1 mm³ MP2RAGE resolution and corrected for B1 field inhomogeneity using the N4 algorithm implemented in ANTs (Tustison et al., 2010). To reconstruct the whole brain tractograms, we generated 3 million streamlines using the iFOD2 algorithm with anatomical priors (R. E. Smith et al., 2012) on the fiber orientation distributions estimated with multi-shell multi-tissue constrained spherical deconvolution (Jeurissen et al., 2014), seeding from the gray matter-white matter interface and propagating the streamlines with the backtrack option using a cut-off value of 0.05 and a maximum angle of 30°. To reduce the incidence of false positives (Buchanan et al., 2020, Campbell and Pike, 2014, Maier-Hein et al., 2017, Zalesky et al., 2016), we set the power parameter of iFOD2 to 3, as in (Bosticardo et al., 2021).

WMLs were semi-automatically segmented using an in-house deep learning-based method (La Rosa et al., 2020), followed by manual correction. Lesion-filled MP2RAGE images were used to segment cortical and subcortical structures into 85 regions of interest (Desikan et al., 2006, Iglesias et al., 2015) using FreeSurfer 6.0 (Fischl, 2012). We employed the boundary-based linear registration tool implemented in FSL (Jenkinson et al., 2002) to register lesion and volumetric segmentation masks to the diffusion space.

Quantitative susceptibility maps (QSM) were reconstructed from 3D EPI data by unwrapping phase, removing the background field through the Projection onto Dipole Fields algorithm, and using the morphology-enabled dipole inversion algorithm to compute the susceptibility from the local field as in (Liu et al., 2012). Possibly remyelinated lesions were identified as those appearing iso-intense on QSM relative to the surrounding normal-appearing white matter, based on the classification described in (Rahmanzadeh et al., 2022). The absence of susceptibility changes in these lesions suggests limited iron accumulation or chronic inflammatory activity, which may be indicative of a remyelinated state.

MTsat maps were computed from the MT-weighted (MTw), proton density-weighted (PDw), and T1w images and corrected for radiofrequency (RF) inhomogeneities using B1 maps, following the procedure described by Helms et al. in (Helms et al., 2008a, Helms et al., 2008b; Helms & Dechent, 2009). The processing was performed using the hMRI Toolbox (Tabelow et al., 2019), which is available at https://github.com/hMRI-group/hMRI-toolbox.

MVF maps were estimated with the following formula: $\mathit{MVF}=\alpha \ast MTsat$, where MTsat represents the portion of free water saturated during a single MT pulse. The calibration constant α was determined according to the procedure described in (Mohammadi et al., 2015).

As the final step of our pipeline, we applied the extended multi-compartment MySD approach (Bosticardo et al., 2025) (available at https://github.com/daducci/COMMIT/wiki/Multicompartment-model-for-pathological-connectomes), using as input the lesion mask segmented from the FLAIR image, the reconstructed tractogram, and the MVF map.

From the resulting TIME map, we derived a single value for each subject by averaging the TIME across all white matter lesions. The same procedure was applied to obtain reference values by averaging MVF within the lesions directly.

2.3. Statistical analysis

We conducted two types of statistical analyses. The first was a baseline cross-sectional analysis aimed at comparing the clinical relevance of TIME with that of the values assessed using raw MVF. The second was a longitudinal analysis, in which we examined changes in both TIME and raw MVF values over time.

As part of the baseline evaluation, we performed two different analysis: (i) we investigated the association between neurological disability and myelin content in WMLs. To test this, we used a robust linear model with EDSS as the dependent variable, myelin damage (assessed either by TIME or directly by MVF) as the independent variable, and sex, age, and disease duration as covariates; and (ii) we investigated the association between neurological disability and the ratio of lesion damage in remyelinated lesions to total lesion damage across all WMLs. This was assessed with a robust linear model using EDSS as the dependent variable and, as the independent variable, the ratio between values obtained from TIME or directly from MVF within remyelinated lesions and those across all WMLs identified on FLAIR.

Following this, we conducted a longitudinal analysis to address two primary questions:

(i) First, we investigated whether longitudinal changes in TIME and in the reference metric within all WMLs over the 2-year follow-up period were associated with baseline EDSS severity. We used a robust general linear model with baseline EDSS as the dependent variable and the annual percentage change in myelin content within WMLs as the independent variable.

(ii) Second, we examined the association between the annual percentage change in myelin content and the occurrence of disability worsening during follow-up, as previously defined by (Lorscheider et al., 2016). Specifically, disability worsening was defined as an increase in EDSS of at least 1.5 points for subjects with a baseline EDSS of 0, at least 1.0 points for those with a baseline EDSS between 0.5 and 5.5, and at least 0.5 points for subjects with a baseline EDSS greater than 5.5. We compared the rate of annual percentage change in myelin content within WMLs between patients with and without disability worsening using the Wilcoxon test. To account for multiple testing, p-values were adjusted using the Bonferroni correction within pre-specified families of hypotheses (i.e., TIME vs MVF within each analysis). Both unadjusted and Bonferroni-adjusted p-values are reported in the Results section.

Fig. 2 graphically illustrates the methodology flowchart used.

Fig. 2.

The image shows a graphic illustration of the pipeline. The patients underwent the MRI scanner twice, two years apart. For each time point, we generated a whole-brain tractogram and segmented the white matter lesions, which were subsequently manually corrected. The lesion mask, the tractograms, and the myelin-sensitive map were used as input to the MySD framework sensitive to lesions, which returned two maps to us. One map shows the bundle-specific myelin content estimated by the model, and a second map provides the values of tractography-informed myelin estimate (TIME), in which the myelin lost by a bundle of neurons in the area affected by the lesion is estimated. Afterward, TIME values in the lesions and the myelin content in MVF maps are compared and investigated for their correlations with the EDSS clinical scale in the whole WM. In addition, the correlation between the TIME values and the patient’s clinical worsening over the years is estimated, and the prediction of myelin damage is compared to the baseline EDSS value. TP: Time Point, MySD: Myelin Streamline Decomposition, MVF: Myelin Volume Fraction, TIME: Tractography-Informed Myelin Estimate, WML: White Matter Lesion.

3. Results

3.1. Cross-sectional analyses at baseline

As shown in Fig. 3, the myelin content within WMLs, as assessed with TIME, was associated with the EDSS (p = 0.015, Bonferroni-adjusted p = 0.031, model R² = 0.51, β = 0.14). In contrast, no significant association was found between myelin content within WMLs as evaluated with MVF and the EDSS (p = 0.220, Bonferroni-adjusted p = 0.44, model R² = 0.50, β = -0.07). The results are summarized in Table 2.

Fig. 3.

Correlation between the EDSS adjusted for disease duration, age, and sex, and the values in the WMLs computed by averaging the MVF values on the left and by Tractography-informed myelin loss values on the right. EDSS: Expanded Disability Status Scale, MVF: Myelin Volume Fraction, TIME: Tractography-Informed Myelin Estimate, WML: White Matter Lesion.

Table 2.

Results of the robust linear model used to test hypothesis H-0: TIME/myelin values in WM lesions are not correlated with the EDSS. In both cases, the model's outcome is the EDSS, and it was adjusted for age, sex, and disease duration. In the upper part of the table, we present the model results using the TIME map in WM lesions calculated using Multi-compartment MySD. In the lower part, we report the results of the model where the independent variable corresponds to MVF values in WM lesions calculated from the MTsat map. Significance levels were denoted with asterisks according to the following convention:' ***' for 0.001, '**' for 0.01, and '*' for 0.05.

Multi-compartment MySD	Estimate	Std. Error	t-value	p-value
(Intercept)	−1.492472	0.571710	−2.611	0.00999 **
TIME	23.776318	9.703435	2.450	0.01546 *
Disease Duration	0.066497	0.013142	5.060	1.25e-06 ***
Age	0.064624	0.009525	6.785	2.75e-10 ***
Sex (male)	−0.181879	0.241227	−0.754	0.45209
Multiple R2: 0.5281 Adjusted R2: 0.515
MTsat-MVF	Estimate	Std. Error	t-value	p-value
(Intercept)	0.817617	1.171456	0.698	0.486
MVF	−4.256031	3.453185	−1.232	0.220
Disease Duration	0.066006	0.013351	4.944	2.09e-06 ***
Age	0.063749	0.009656	6.602	7.13e-10 ***
Sex (male)	−0.158479	0.242954	−0.652	0.515
Multiple R2: 0.5137 Adjusted R2: 0.5003

The remyelination ratio, defined as the ratio between values derived either by TIME or by the reference method within iso-/hypo-intense QSM lesions and those within all WMLs, was not significantly associated with EDSS (p = 0.662 and p = 0.866, respectively). These findings are presented in Table 3 and visualized in Fig. 4.

Table 3.

Results of the robust linear model used to test hypothesis H-2: The ratio between the values inside the remyelinated lesions and the values inside the WM lesions are unrelated to the EDSS. The model's outcome is the EDSS, which was adjusted for age, sex, and disease duration. In the upper part of the table, we present the model results using the TIME map computed using Multi-compartment MySD. In the lower part, we report the results of the model where the independent variable corresponds to MVF values calculated from the MTsat map. Significance levels were denoted with asterisks according to the following convention:' ***' for 0.001, '**' for 0.01, and '*' for 0.05.

Multi-compartment MySD	Estimate	Std. Error	t-value	p-value
(Intercept)	−1.06802	1.01298	−1.054	0.297617
TIME_ratio	0.13832	0.31419	0.440	0.661971
DiseaseDuration	0.07279	0.02401	3.032	0.004107 **
Age	0.08035	0.02109	3.810	0.000437 ***
Sexm	−0.28946	0.47904	−0.604	0.548851
Multiple R2: 0.5135 Adjusted R2: 0.4682
MTsat-MVF	Estimate	Std. Error	t-value	p-value
(Intercept)	−1.39876	1.53791	−0.910	0.368145
MVF_ratio	0.10637	0.62866	0.169	0.866425
DiseaseDuration	0.07291	0.02421	3.012	0.004339 **
Age	0.07929	0.02143	3.699	0.000609 ***
Sexm	−0.26689	0.48264	−0.553	0.583143
Multiple R2: 0.5107 Adjusted R2: 0.4652

Fig. 4.

Correlation between the EDSS estimated at baseline adjusted for disease duration, age, and sex, and the remyelinated lesion ratio in 1) average MVF values within WMLs (left) and 2) tractography-based myelin damage in WMLs (right). EDSS_BL: Expanded Disability Status Scale at Baseline, MVF: Myelin Volume Fraction, TIME: Tractography-Informed Myelin Estimate, WML: White Matter Lesion.

3.2. Longitudinal analyses

Patients with a greater percentage change as assessed using TIME, reflecting a higher lesion contribution fitted by the model, showed significantly higher EDSS scores at follow-up. Specifically, each 1 % increase in percentage change was associated with a + 0.54 % increase in expected EDSS (β = 0.00538 ± 0.00066, 95 % CI = [0.00408; 0.00667], p < 0.001, Bonferroni-adjusted p < 0.001, R² = 0.37).

In contrast, the percentage change assessed using raw MVF was not significantly associated with EDSS scores at follow-up (–0.36 % change, β = –0.00356 ± 0.00220, 95 % CI = [-0.0079 – 0.0007], p = 0.105, Bonferroni-adjusted p = 0.21, R² = 0.06). Fig. 5 shows the corresponding associations for both TIME and MVF. Table 4 reports the association between baseline EDSS and the percentage changes assessed using TIME and raw MVF.

Fig. 5.

Correlation between the EDSS estimated at baseline and the annual percentage change in 1) average MVF values within WMLs (left) and 2) tractography-based myelin damage in WMLs (right). EDSS_BL: Expanded Disability Status Scale at Baseline, MVF: Myelin Volume Fraction, TIME: Tractography-Informed Myelin Estimate, WML: White Matter Lesion.

Table 4.

Results of the robust general linear model used to evaluate if the EDSS estimated at baseline predicts the percentage annual change in the myelin damage values inside the WML computed using MVF and TIME values. The model's outcome is the EDSS, while the dependent variable is the annual percentage change inside WMLs. The β coefficients are on the log scale, while OR and 95% CI are reported on the original scale (i.e., multiplicative effects on EDSS). Significance levels were denoted with asterisks according to the following convention:' ***' for 0.001, '**' for 0.01, and '*' for 0.05.

Model	Predictor	β	SE	p-value	95 % CI	R2
Multi-compartment MySD	Intercept	1.139	0.0436	<0.001***	–	0.369
	Percentage change	0.00538	0.00066	<0.001***	[0.00408–0.00667]
MTsat-MVF	Intercept	1.365	0.0379	<0.001***	–	0.06
	Percentage change	−0.00356	0.00220	0.105	[-0.0079 – 0.0007]

Patients who experienced disability worsening during the follow-up period showed significantly greater percentage increases in TIME compared to clinically stable patients (p = 0.044, Bonferroni-adjusted p = 0.088, W = 496). A similar trend was observed when using the percentage change as evaluated with MVF, although the difference did not reach statistical significance (p = 0.059, Bonferroni-adjusted p = 0.118, W = 510) (Fig. 6).

Fig. 6.

In the plot, the MVF average and tractography-informed myelin loss annual percentage changes in WMLs are reported in the two groups of MS patients: worsening and not worsening. MVF: Myelin Volume Fraction, TIME: Tractography-Informed Myelin Estimate, WML: White Matter Lesion.

4. Discussion

In this study, we proposed a tractography-informed myelin estimation framework, referred to as TIME, which quantifies voxel-wise myelin loss within WMLs relative to the unaffected portions of the same tract. This approach builds on microstructure-informed tractography models (Daducci et al., 2015, Daducci et al., 2016, Zhang et al., 2022), which assume constant microstructural properties along axonal bundles represented by streamlines. However, in the presence of focal pathology, this assumption no longer holds. To address this, the model was recently extended to allow for the relaxation of the constant-microstructure assumption specifically within lesional voxels (Bosticardo et al., 2025). When the compartment representing the myelin-sensitive signal is insufficient to fit the observed data, a dedicated lesion compartment is introduced to help model the deviation from microstructural constancy in the lesioned region. As a result, the output map provides an estimate of myelin damage that reflects the proportion of the lesion compartment required to model local signal deviations along the tract. In this study, we tested this modeling approach and found that it improves sensitivity to clinically meaningful alterations in WMLs, compared to conventional methods based on simple signal averaging within lesions, which do not incorporate anatomical or tract-specific context.

By applying TIME to an MS cohort using MTsat-derived MVF maps, we demonstrated the value of this framework in both cross-sectional and longitudinal settings. Specifically, TIME values within WMLs were significantly associated with clinical disability, as measured by EDSS, whereas conventional MVF averaging within WMLs showed no such association. Moreover, longitudinally, higher baseline disability predicted a faster increase in TIME values over time, and this rate of increase was in turn associated with clinical worsening. These findings suggest that TIME captures focal demyelination in a way that is both biologically grounded and clinically relevant.

Our findings refine and extend a growing body of work investigating white matter damage in MS using microstructural imaging. A large portion of the existing literature has focused on alterations in normal-appearing white matter, typically using global or regional diffusion or magnetization metrics. For instance, Bezukladova et al. (2020) showed that lower fractional anisotropy in the normal-appearing white matter correlates with greater disability, supporting the idea that diffuse microstructural damage contributes to clinical impairment. Similarly, York et al. (2022) reported a negative association between magnetization transfer ratio in the normal-appearing white matter and EDSS. While informative, these metrics are not specific to myelin content and may reflect a combination of demyelination, axonal loss, edema, and inflammation. In addition, their global or regional averaging across normal-appearing white matter may mask subtle but functionally relevant focal damage.

In parallel, several studies have directly examined white matter lesions. Kolind et al. (2012) showed that patients with progressive MS exhibit widespread reductions in myelin water fraction in both normal-appearing white matter and WMLs compared to healthy controls, with myelin water fraction reductions correlating with disability. Margoni et al. (2022) found that T1/T2 ratio values were significantly lower in T2-hyperintense lesions across all MS phenotypes, with the lowest values observed in secondary progressive MS compared to RRMS. They also reported associations between lower T1/T2 values in T2 hyperintensity lesions and longer disease duration, higher EDSS, and lower brain volume. While these studies highlight the clinical relevance of lesional tissue properties, their reliance on comparisons with normative atlases or group-level differences may obscure individual variability.

However, not all lesion-based analyses have identified a clear association between myelin loss and clinical disability. For example, (Oreja-Guevara et al., 2006) reported no significant association between magnetization transfer ratio values averaged across lesions and EDSS over an 18-month follow-up. This discrepancy may reflect the limited specificity of magnetization transfer ratio to myelin content (York, Meijboom, et al., 2022), or the influence of more complex pathological mechanisms, such as coexisting axonal loss, inflammation, or incomplete repair, that are not captured by simple averaging approaches. Supporting this, Rahmanzadeh et al. (2022) classified more than 2800 WMLs based on their QSM signal into iso-, hypo-, and hyperintense lesions, as well as paramagnetic rim lesions, and showed that these lesion types differ substantially in their associated myelin water fraction and neurite density index. Notably, iso-intense lesions exhibited myelin water fraction values comparable to those of normal-appearing white matter and healthy controls, whereas paramagnetic rim lesions and hyperintense lesions were associated with more severe myelin and axonal damage.

Our approach, rather than relying on global references or normative atlases, estimates myelin loss at each lesional voxel by modeling the lesion-related signal contribution that needs to be added to account for the deviation from the expected microstructure within the voxels of the healthy portion of the same tract. This allows each lesional voxel to be interpreted in the context of the healthy segment of the bundle crossing the lesion, preserving anatomical specificity and accommodating inter-subject variability. Although lesion-level values are ultimately summarized by averaging voxel-wise TIME estimates, these estimates reflect localized deviations from tract integrity rather than raw intensity values. In contrast to approaches based on normative comparisons or region-wide averaging, TIME provides a biologically grounded and anatomically contextualized measure of lesion severity. This likely explains its superior sensitivity to neurodegeneration, as assessed through EDSS, both cross-sectionally and longitudinally, compared to standard MVF averaging.

Moreover, as mentioned before, the TIME framework builds on a previously published model (Bosticardo et al., 2025) that can be flexibly applied using various additive imaging contrasts, including diffusion-based metrics such as the neurite density index. Importantly, the input map must reflect an additive biological quantity; therefore, non-additive metrics such as fractional anisotropy are not suitable for this model. While diffusion-derived inputs may be particularly suitable in conditions characterized by axonal loss or edema, in the case of MS, we specifically chose a myelin-sensitive map due to the central role of demyelination in WMLs. The model’s flexibility allows it to be adapted to different pathologies, but its input metric should reflect the dominant biological process under investigation. In the context of MS, this means prioritizing metrics sensitive to myelin.

We also explored whether TIME could reflect remyelination by comparing values in lesions classified as hypo- or iso-intense on QSM, often interpreted as remyelinated or less damaged, relative to the full WML lesion set. However, this ratio did not significantly correlate with EDSS. This may indicate that relative signal preservation in selected lesions does not necessarily translate into functional benefit, or that other components of damage (e.g., axonal loss) have a greater impact on clinical scores. It is also possible that the effect size is subtle and difficult to detect with current sample sizes and outcome measures. Interestingly, Rahmanzadeh et al. (2022) also found that myelin water fraction in iso-intense lesions was not significantly different from normal-appearing white matter, but these lesions still differed in neurite integrity, further suggesting that multimodal approaches may be needed to fully capture clinically relevant repair.

Crucially, although this study focused on MS, the TIME framework inherits the flexibility of the underlying model described in (Bosticardo et al., 2025), and can therefore be applied to other focal white matter pathologies. TIME is not disease-specific and may be used to investigate lesion-related mechanisms in a variety of neurological conditions. The only requirement is the availability of a pre-segmented lesion mask and a microstructural map that aligns with the biological question under investigation. Since TIME operates at the voxel level, it can capture spatial gradients and intra-lesional variability − features increasingly recognized as important for understanding lesion evolution. For instance, partial remyelination or rim-associated activity may produce subtle signal gradients that TIME may be inherently capable of detecting. Future developments could further strengthen these capabilities, supporting a more refined characterization of lesion staging and progression.

Some limitations of this work should be acknowledged. First, the method depends on the accuracy of lesion masks, which may vary depending on the segmentation approach (Styner et al., 2008) and can propagate errors into the TIME estimates. Second, the framework relies on accurate tractography reconstruction, which can be impacted by various challenges, such as crossing fibers, differences in anatomy, and imaging artifacts (Daducci et al., 2016, Maier-Hein et al., 2017, Rheault et al., 2025). Although we utilized validated pipelines, these issues remain significant challenges in the field. Third, the longitudinal analysis in this study was limited to two time points, which constrains our ability to model non-linear trajectories of demyelination and remyelination. Future studies with more frequent and longer follow-up intervals would be better suited to characterizing the temporal dynamics of myelin integrity. Additionally, integrating TIME with functional outcomes, such as cognitive performance or motor testing, could provide deeper insights into the relationship between microstructural damage and clinical manifestations.

Collectively, these findings suggest that TIME provides a tract-specific, biologically informed, and clinically relevant approach to quantifying lesional myelin loss. By modeling damage relative to unaffected tissue within the same anatomical bundle, it increases sensitivity to focal pathology and overcomes the limitations of standard voxel-wise or global averaging techniques. TIME emerges as a promising biomarker for monitoring demyelination, assessing remyelination, and advancing our understanding of MS pathophysiology. Moreover, its flexibility and generalizability support its potential application across a broad spectrum of focal white matter disorders.

5. Conclusion

This study assessed the clinical relevance and sensitivity of a new tractography-informed map of myelin damage, referred to as TIME, derived using the multi-compartment MySD framework. TIME estimates myelin loss by comparing lesioned and non-lesioned portions of the same white matter bundle within each subject, providing an individualized measure of how myelin deteriorates or repairs over time.

Our findings indicate that although TIME is derived from MVF, it provides clinically meaningful insights beyond those offered by MVF alone. This may improve our understanding of disease mechanisms and aid in the development of more sensitive biomarkers for tracking demyelination and evaluating treatment effects in MS.

CRediT authorship contribution statement

Sara Bosticardo: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Matteo Battocchio: Writing – review & editing, Software, Methodology, Investigation, Conceptualization. Mario Ocampo-Pineda: Writing – review & editing, Visualization, Data curation. Alessandro Cagol: Writing – review & editing, Visualization, Investigation, Data curation, Conceptualization. Po- Jui Lu: Writing – review & editing, Investigation, Data curation. Esther Ruberte: Writing – review & editing, Data curation. Nina De Oliveira S. Siebenborn: Writing – review & editing, Data curation. Xinjie Chen: Writing – review & editing, Data curation. Lester Melie-Garcia: Writing – review & editing, Validation, Investigation, Data curation. Matthias Weigel: Writing – review & editing, Data curation. Ludwig Kappos: Writing – review & editing, Funding acquisition. Jens Kuhle: Writing – review & editing, Funding acquisition. Alessandro Daducci: Writing – review & editing, Writing – original draft, Supervision, Investigation, Conceptualization. Cristina Granziera: Writing – review & editing, Writing – original draft, Supervision, Project administration, Investigation, Funding acquisition, Conceptualization.

Funding

The work was supported by the Swiss National Science Foundation (SNSF) grants PP00P3\_206151, 32003BE\_232739.

Declaration of Competing Interest

S. Bosticardo has nothing to disclose. M. Battocchio has nothing to disclose. M. Ocampo-Pineda has nothing to disclose. A. Cagol has received speaker honoraria from Novartis and Roche. P.-J. Lu has nothing to disclose. E. Ruberte has nothing to disclose. N. de Oliveira Siebenborn has nothing to disclose. X. Chen has nothing to disclose. L. Melie Garcia has nothing to disclose. M. Weigel has nothing to disclose. L. Kappos has received no personal compensation. His institutions (University Hospital Basel/Foundation Clinical Neuroimmunology and Neuroscience Basel) have received and used exclusively for research support: payments for steering committee and advisory board participation, consultancy services, and participation in educational activities from: Actelion, Bayer, BMS, df-mp Molnia & Pohlmann, Celgene, Eli Lilly, EMD Serono, Genentech, Glaxo Smith Kline, Janssen, Japan Tobacco, Merck, MH Consulting, Minoryx, Novartis, F. Hoffmann-La Roche Ltd, Senda Biosciences Inc., Sanofi, Santhera, Shionogi BV, TG Therapeutics, and Wellmera, and license fees for Neurostatus-UHB products; grants from Novartis, Innosuisse, and Roche. J. Kuhle received speaker fees, research support, travel support, and/or served on advisory boards by Swiss MS Society, Swiss National Research Foundation (320030_189140/1), University of Basel, Progressive MS Alliance, Bayer, Biogen, Celgene, Merck, Novartis, Octave Bioscience, Roche, Sanofi. A. Daducci has nothing to disclose. C. Granziera The University Hospital Basel (USB), as the employer of C.G., has received the following fees which were used exclusively for research support: (i) advisory boards and consultancy fees from Actelion, Novartis, Genzyme-Sanofi, GeNeuro, Hoffmann La Roche and Siemens; (ii) speaker fees from Biogen, Hoffmann La Roche, Teva, Novartis, Merck, Jannsen Pharmaceuticals and Genzyme-Sanofi; (iii) research grants: Biogen, Genzyme Sanofi, Hoffmann La Roche.

Data availability

Data will be made available on request.