Widespread white matter microstructural abnormalities in treatment‑naïve patients with first‑episode schizophrenia revealed by fixel-based analysis

Abstract

Fixel-based analysis (FBA) is an advanced diffusion imaging method that enables the direct estimation of white matter microstructural properties beyond the limitations of traditional diffusion tensor imaging (DTI). Despite its potential, FBA has been rarely applied in schizophrenia research, and its value in providing complementary information to conventional tensor-based approaches remains to be fully established. In this study, we investigated white matter abnormalities of treatment-naïve, first-episode schizophrenia (FES) patients using both FBA and tensor-based method, and examined the concordance between the two approaches to better characterize the nature of white matter pathology in early SZ. MRI data were acquired from 94 treatment-naïve FES patients and 114 healthy controls (HCs). Fractional anisotropy (FA) and mean diffusivity (MD) were calculated using a conventional tensor-based method. In parallel, fibre density (FD), fibre-bundle cross-section (FC), and their combined metric (FDC) were estimate with FBA. White matter was segmented into 72 anatomically defined tracts based on fibre tracking. Between-group comparisons were conducted using a multivariate general linear model (GLM) to assess differences across diffusion metrics. Using the tensor-based method, six white matter tracts exhibited significantly altered FA, while 34 tracts showed significantly increased MD in FES patients compared to HCs (all t-values > 2.34 or t-values < −2.36, all FDR-p < 0.05). In contrast, FBA revealed more widespread abnormalities: 46 tracts showed significantly reduced FD, 29 tracts showed significantly reduced FC, and 52 tracts showed significantly reduced FDC (all t-values < −2.29, all FDR-p < 0.05). Notably, all tracts with significantly reduced FC metrics also demonstrated corresponding FDC reductions. No significant correlation was observed between any diffusion metrics and clinical characteristics (all FDR-p ˃ 0.05). This study highlights the remarkable advantages of the FBA in detecting WM microstructural abnormalities in individuals with FES.

Introduction

Schizophrenia (SZ) is a severe psychiatric disorder associated with substantial personal and societal burden, as well as low quality of life, and comorbidities [1–4]. Neuroimaging studies have provided valuable insights into its underlying pathophysiological mechanisms, particularly through the investigation of white matter (WM) abnormalities [5]. Diffusion tensor imaging (DTI) has been widely employed to assess WM microstructure [6], with fractional anisotropy (FA) and mean diffusivity (MD) being the most commonly used metrics. FA reflects the degree of directional water diffusion, while MD represents the overall magnitude of diffusion. Both serve as indirect indicators of WM microstructural integrity [7]. Previous large-scale studies have reported widespread reductions in FA and increases in MD across most WM tracts in individuals with chronic SZ compared to healthy controls (HCs) [8]. Abnormal FA and MD have also been observed in early-stage, drug-naïve SZ patients. However, findings have been inconsistent across studies [9–12]. While DTI remains a widely used and informative technique, it has notable limitations. Specifically, the tensor-based model may be inadequate for accurately characterizing complex fibre architecture, particularly at regions where multiple fibre populations intersect [13]. Addressing these methodological limitations is essential for advancing our understanding of WM abnormalities in SZ.

The fixel-based analysis (FBA) framework was proposed to overcome the limitations inherent in traditional tensor-based diffusion model. A fixel, a term derived from “fibre” and “voxel”, represents a specific fibre population within a voxel, identified by calculating the WM fibre orientation distributions (FODs) using constrained spherical deconvolution (CSD). Unlike the tensor model, which assumes a single dominant fibre direction per voxel, the FBA approach allows for the identification of multiple distinct fibre populations within the same voxel [14]. FBA provides three metrics including fibre density (FD), fibre cross-section (FC), and their combined measure, fibre density and cross-section (FDC). FD reflects the intra-axonal volume fraction, indicating microstructural density along a specific orientation. FC captures the macroscopic cross-sectional size of a fibre bundle, reflecting morphological differences such as atrophy or expansion. FDC, as the product of FD and FC, integrates both micro- and macrostructural properties of white matter bundles [15]. Compared to conventional tensor-based approaches, FBA offers anatomically specific, fibre-orientation-resolved insights into WM structure, enabling more precise characterization of complex fibre architectures [16].

The FBA framework has increasingly been adopted in diffusion Magnetic Resonance Imaging (dMRI) studies and detected a greater number of WM tracts with alterations than the traditional tensor-based model. For example, in Alzheimer’s disease research, FBA-derived metrics showed higher specificity in identifying tau-related WM alterations compared to FA and MD metrics, particularly during the early stages of the disease. These alterations were primarily localized in the cingulum, uncinate fasciculus (UF), and anterior thalamic radiations (ATR) [17]. Similarly, in a study of older adults with metabolic syndrome, significant reductions in FBA-derived metrics were observed in multiple WM tracts, including the corticospinal tract (CST), superior cerebellar peduncle (SCP), and middle cerebellar peduncle (MCP), whereas conventional tensor-based metrics failed to reveal any abnormalities [18]. The utility of FBA also extends to psychiatric disorders. It has been successfully applied in studies on attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), major depressive disorder (MDD), and chronic cannabis use disorder [19–23]. To date, a limited number of studies have utilized FBA to investigate WM abnormalities in SZ [24–26].

Stämpfli et al. reported reduced FD in the thalamic radiation (TR), superior longitudinal fasciculus (SLF), corpus callosum (CC), and CST in individuals with chronic SZ, while no significant differences in FA were observed [25]. This suggests that FBA may be more specific in detecting WM microstructural abnormalities than the traditional tensor-based method. However, the small sample size (20 SZ patients and 25 HCs) limits the generalizability of the findings [25]. Grazioli et al. found reduced FD in the postcentral gyrus and posterior CC among medicated SZ individuals, while FC remained unchanged [24]. The heterogeneity in disease status and the mitigating effects of antipsychotic treatment may have confounded these results [24]. More recently, Kristensen et al. examined 86 antipsychotic-naïve first-episode SZ (FES) patients and observed reduced FD in the CC. However, when analyses were restricted to a subset of substance-free FES individuals, the significance of these findings disappeared, likely due to reduced statistical power following sample size reduction [26]. To date, there remains a notable paucity of large-sample FBA studies specifically targeting treatment-naïve FES individuals.

In this study, to address the limitations of previous study, we applied both the FBA method and the tensor-based method to a large sample of treatment-naïve individuals with FES. By using the two methods, this study aimed to provide a comprehensive characterization of white matter alterations in early SZ. Furthermore, we explore the relationship between the FBA-based metrics and tensor-based metrics. Additionally, the FBA method is sensitive in crossing-fibre regions, we hypothesized that the FBA method would detect more extensive and fibre-specific alterations than conventional tensor-based metrics.

Methods and materials

Participants

Treatment-naïve patients with FES and HCs were recruited from the Mental Health Center of West China Hospital, Sichuan University. All participants were right-handed and Han Chinese. FES patients were interviewed by trained psychiatrists using the Structured Clinical Interview for the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), patient version (SCID-P). HCs were assessed using the non-patient version of the SCID (SCID-NP) to confirm the absence of current or past psychiatric disorders. All FES participants met the diagnostic criteria for SZ according to the DSM-IV and were treatment-naïve at the time of MRI scanning. Exclusion criteria for both groups included a history of alcohol or substance abuse, major neurological or severe physical illness, and any contraindication to MRI.

This study was approved by the Institutional Review Board of West China Hospital, Sichuan University (approval no.2017-131), and was conducted in accordance with the principles outlined in the Helsinki Declaration. Written informed consent was obtained from all participants or their legal guardians after a full explanation of the study procedures and objectives.

Clinical assessment

Clinical symptoms in FES individuals were evaluated using the Positive and Negative Symptom Rating Scale (PANSS) [27], while overall functioning was evaluated with the Global Assessment of Functioning (GAF) scale [28]. Intelligence quotient (IQ) for all participants was measured using the short version of the Wechsler Adult Intelligence Scale-Revised in China (WAIS-RC) [29]. This version includes seven subtests including Information, Arithmetic, Digital Symbol, Digital Span, Block Design, Picture Completion, and Similarities.

Image acquisition

MRI scanning was performed using a Philips 3.0 T scanner (Achieva, TX, Best, Amsterdam, The Netherlands) with an eight-channel phased-array head coil. Foam padding and headphones were used to minimize head motion and reduce the interference of scanner noise during the scanning process. Diffusion weighted imaging (DWI) images were acquired using a single-shot spin-echo echo-planar imaging sequence (EPI) with the following parameters: diffusion gradient directions = 32 with b-value = 1000 s/mm², a single unweighted volume with b-value = 0 s/mm², repetition time (TR) = 10, 407 ms, echo time (TE) = 92 ms, flip angle = 90°, slice thickness = 2.0 mm, SENSE factor 2 in the anterior-posterior direction, field of view (FOV) = 256 × 256 mm², voxel size = 2 × 2 × 2 mm³, matrix size = 128 × 128, EPI factor = 67, and slices = 75. The T1-weighted (T1w) images were acquired by a magnetization-prepared rapid gradient-echo sequence (MPRAGE): TR = 8.1 ms, TE = 3.7 ms, inversion time = 1072.4 ms, flip angle = 7°, slice thickness = 1 mm (no slice gap), 188 axial slices, matrix size = 256 × 256, FOV = 256 × 256 mm², voxel size = 1 × 1 × 1 mm³.

Image preprocessing

The T1w image was corrected for intensity non-uniformity (INU) using N4BiasFieldCorrection [30] (ANTs 2.4.3) to obtain the anatomical reference image. This reference image was then reoriented into Anterior Commissure-Posterior Commissure (AC-PC) alignment using a 6-degree-of-freedom (6-DOF) transform which is extracted from a full affine registration to the MNI152NLin2009cAsym template. A full nonlinear registration to the template was performed in AC-PC space using symmetric nonlinear registration (SyN) via antsRegistration. Brain extraction was carried out using SynthStrip [31], and automated segmentation was conducted using SynthSeg [32], from FreeSurfer (version 7.3.1).

Marchenko-Pastur PCA (MP-PCA) denoising was applied using MRtrix3’s “dwidenoise” with a 5-voxel window [33]. B1 field inhomogeneity was corrected using MRtrix3’s “dwibiascorrect” with the N4 algorithm [30] after the images were resampled. Head motion and eddy current-induced distortions were corrected with FSL’s eddy tool (version 6.0.5) [34]. Eddy’s outlier replacement method was applied [35], in which data were grouped by slice and only those slices containing at least 250 intracerebral voxels were retained. Slices deviating more than 4 standard deviations from the predicted values had their data replaced with imputed values.

FBA processing

Fixel-based processing steps were carried out using MRtrix3Tissue, a specific version of MRtrix3 for single-shell 3-tissue CSD (SS3T-CSD) (https://3tissue.github.io) [36]. Preprocessed DWI images were converted to MRtrix3 format (“.mif”) using “mrconvert” and were processed according to the MRtrix3 fixel-based analysis steps (www.MRtrix.readthedocs.io) [15]. For each subject, a single unique set of 3-tissue response functions representing single fibre WM, grey matter (GM) and cerebrospinal fluid (CSF) were generated. Then, the average of all subject response functions for each tissue type were computed [37]. Preprocessed images were upsampled to an isotropic voxel size of 1.25 mm to increase anatomical contrast. A whole brain mask was computed from the upsampled images. SS3T-CSD was performed for each subject using the set of average response functions to obtain WM-like FOD estimation [37]. Joint bias field correction and global intensity normalization were performed using “mtnormalise”. A study-specific unbiased FOD template was generated from 60 subjects (30 patients and 30 controls matched on age and gender). An upsampled FA template was also generated in the same step. Each subject’s WM FOD image was registered to the FOD template. To ensure subsequent analysis is performed in voxels that contain data from all subjects, all subject masks were warped into template space and the FOD template mask was computed as the intersection of all warped masks. Subjects’ FOD images were then warped into template space without FOD reorientation.

The FA template was thresholded at a value of 0.2 to create a voxel mask to rule out GM and CSF. This voxel mask was then multiplied with the FOD template mask to generate the final analysis mask. The fixel mask was generated from the final analysis mask and a peak value threshold of 0.08. Subjects’ FOD images without reorientation were segmented into FD-based fixels within each voxel and reoriented to the template space. The FD images for each subject were computed by assigning each fixel in reoriented subject space to the corresponding fixel in template space. The FC metric based on fixel mask was computed from the warps generated during registration, and log (FC) was calculated for analysis to ensure data are normally distributed. A combined metric of FDC metric was computed as the product of FD and FC. Then whole-brain fibre tractography was performed on the FOD template. The tractography result was used to generate the fixel-fixel connectivity matrix, which was used to smooth the FD, log (FC) and FDC data along WM tracks directions.

DTI processing

A tensor model was fitted at each voxel using MRtrix3’s “dwi2tensor” function to estimate the diffusion tensor, from which FA and MD images were generated. The FA images were used to generate the study specific FA template during the FOD template building step. Then, the FA and MD images were registered to the study specific template space using the same warp matrix as the FOD images.

WM tracts segmentation

We used TractSeg (v2.9) to generate fibre tracking of 72 anatomically well described tracts on the study-specific FOD template [38]. The WM tracts including bilateral arcuate fascicle (AF), anterior thalamic radiation (ATR), cingulum (CG), corticospinal tract (CST), middle longitudinal fascicle (MLF), fronto-pontine tract (FPT), fornix (FX), inferior cerebellar peduncle (ICP), inferior occipito-frontal fascicle (IFO), inferior longitudinal fascicle (ILF), optic radiation (OR), pariet-occipital pontine (POPT), superior cerebellar peduncle (SCP), superior longitudinal fascicle Ⅰ (SLF_Ⅰ), superior longitudinal fascicle Ⅱ (SLF_Ⅱ), superior longitudinal fascicle Ⅲ (SLF_Ⅲ), superior Thalamic radiation (STR), uncinate fascicle (UF), thalamo-prefrontal (T_PREF), thalamo-premotor (T_PREM), thalamo-precentral (T_PREC), thalamo-postcentral (T_POSTC), thalamo-parietal (T_PAR), thalamo-occipital (T_OCC), striato-fronto-orbital (ST_FO), striato-prefrontal (ST_PREF), striato-premotor (ST_PREM), striato-precentral (ST_PREC), striato-postcentral (ST_POSTC), striato-parietal (ST_PAR), and striato-occipital(ST_OCC), commissure anterior (CA), corpus callosum and its 7 segments (all: CC, segments: CC_1-8), and middle cerebellar peduncle (MCP) (Supplementary Fig. S1). Tracking files were converted into fixel images with “tck2fixel” function in MRtrix3, then a binary voxel mask for each WM tract was generated from the corresponding fixel image with “fixel2voxel” function. Mean values of FD, log (FC), and FDC of 72 WM tracts were extracted with their fixel masks. Mean values of FA and MD of 72 WM tracts were extracted with their voxel masks.

Statistical analysis

Clinical data and extracted MRI metrics were analyzed in R software (Version 4.3.1). Prior to group comparisons, all variables were tested for normality using the one-sample Kolmogorov–Smirnov test and visually inspected via quantile-quantile plot (Q-Q plot). Group comparisons of FA, MD, FD, log (FC), and FDC metrics were performed using Multivariate General Linear Model (GLM). Age, gender, and years of education were included as covariates to control for potential confounding effects. Pearson correlation analyses were conducted between the vectors of inter-group t values across all the 72 tracts for each tensor-based metric (FA and MD) and each FBA-based metric (FD, log (FC), and FDC). To assess the clinical relevance of altered MRI metrics, partial correlation analyses were conducted in the FES group to examine associations with clinical measures, including duration of untreated psychosis (DUP), PANSS scores, and cognitive performance, while adjusting for age, gender, and years of education. All results were corrected for multiple comparisons using the false discovery rate (FDR) method, with statistical significance defined as FDR-p < 0.05.

Results

Demographic and clinical characteristics

We recruited 94 drug-naïve FES individuals (47 males) and 142 HCs (41 males) in this study. As shown in Table 1, FES participants were significantly younger and had lower years of education compared to HCs. Moreover, FES individuals exhibited significantly poorer cognitive performance, reflected by lower total IQ, verbal intelligence quotient (VIQ), performance IQ (PIQ), as well as reduced scores on the digital symbol, and digital span subtests of the WAIS-RC.

Table 1.

Sociodemographic and clinical data of all participants.

Characteristic	FES (N = 94)	HCs (N = 142)	t/χ2	p value
Gender (male/female)	47/47	41/101	9.91	0.002
Age (years)	22.44 ± 6.53	24.56 ± 5.55	−2.59	0.010
Education (years)	12.26 ± 2.53	15.22 ± 2.27	−9.13	<0.001
VIQ	96.26 ± 20.47	115.64 ± 13.24	−7.83	<0.001
PIQ	93.9 ± 18.2	113.29 ± 11.66	−8.83	<0.001
WAIS IQ	95.67 ± 17.58	115.94 ± 12.37	−9.35	<0.001
Digital symbol	45.91 ± 16.32	63.17 ± 18.37	−6.99	<0.001
Digital span	13.77 ± 2.42	15.83 ± 2.78	−5.60	<0.001
DUP (month)	6 (1, 12)	–	–	–
PANSS total score	89.46 ± 20.1	–	–	–
PANSS PS score	23.49 ± 5.82	–	–	–
PANSS NS score	22.93 ± 7.71	–	–	–
PANSS GP score	43.04 ± 11.45	–	–	–
GAF	46.99 ± 11.58

Normal values are presented as mean ± standard deviation, non-normal variables are presented as median (P25, P75).

FES first-episode schizophrenia, HCs healthy controls, WAIS Wechsler adult intelligence scale, IQ intelligence quotient, VIQ verbal intelligence quotient, PIQ performance intelligence quotient, DUP duration of untreated psychosis, PANSS the positive and negative symptom scale, PS positive symptom scale, NS negative symptom scale, GP general psychopathology, GAF global assessment function.

Group differences of tensor-based metrics

Statistically significant alterations in FA were observed in 6 WM bundles (Fig. 1A, Supplementary Table S1 and Figure S2). Specifically, FA was reduced in right OR, right T_OCC, and bilateral FX, while FA increased in left ST_PREC and left T_PREC (all t-values > 2.58 or t-values < −2.36, all FDR-p < 0.05). In contrast, MD was significantly increased in 34 bundles (Fig. 1B, Supplementary Table S2 and Figure S3). These bundles included the right AF, right SLF_II, left MLF, CC and its 4 segmentations (CC_2, 3, 6 and 7), as well as bilateral alterations in the ATR, FX, IFO, ILF, OR, ST_FO, ST_OCC, ST_PREF, ST_PREM, T_OCC, T_PAR, T_PREF, and T_PREM (all t-values > 2.34, all FDR-p < 0.05).

Fig. 1. Comparison of tensor-based metrics between groups.

A inter-group comparison of fractional anisotropy; B inter-group comparison of mean diffusivity. The bars represent the t values for inter-group comparisons after controlling for age, gender, and education. White matter with significant differences is marked in red. L left, R right, M middle. *p < 0.05, **p < 0.01, ***p < 0.001.

Group differences of FBA-derived metrics

A statistically significant decrease in FD was observed in 46 WM bundles (Fig. 2A, Supplementary Table S3 and Figure S4). These bundles included the right ICP, left SLF_II, left ST_PREF, left T_POSTC, left T_PREF, left T_PREM, CC and its 5 segmentations (CC_1, 2, 5, 6, and 7), as well as bilateral AF, ATR, CG, FX, IFO, ILF, MLF, OR, POPT, SCP, SLF_III, ST_FO, ST_OCC, ST_PAR, T_OCC, T_PAR, and UF (all t-values < −2.29, all FDR-p < 0.05).

Fig. 2. Comparison of FBA-derived metrics between groups.

A inter-group comparison of fibre density; B inter-group comparison of fibre cross-section (log₁₀-transformed); C inter-group comparison of the product of fibre density and cross-section. The bars represent the t values for inter-group comparisons after controlling for age, gender, and education. White matter with significant differences is marked in red. L left, R right, M middle. *p < 0.05, **p < 0.01, ***p < 0.001.

A statistically significant decrease in log (FC) was found in 29 WM bundles (Fig. 2B, Supplementary Table S4 and Figure S5). These bundles included the right UF, left AF, left CG, 2 segmentations of CC (CC_6 and 7), as well as bilateral IFO, ILF, MLF, OR, POPT, SLF_III, SLF_II, SLF_ I, ST_OCC, ST_PAR, T_OCC, and T_PAR (all t-values < −2.29, all FDR-p < 0.05).

A statistically significant decrease in FDC was found in all bundles that also showed a significant reduction in log (FC) (Fig. 2C, Supplementary Table S5 and Figure S6). Additionally, reduced FDC was identified in 23 other WM bundles, including the right AF, CG, CST, ST_PREC, ST_PREF, and T_PREC, left ATR, ST_PREM, T_PREM, and UF, CC and its 2 segmentations (CC_1, and 2), as well as bilateral FX, ST_FO, ST_POSTC, and T_POSTC. These alterations in FDC partially overlapped with the inter-group differences in FD (all t-values < −2.29, all FDR-p < 0.05).

Correlations of inter-group t values between tensor-based metrics and FBA-derived metrics

Correlations between the inter-group t values of tensor-based metrics and those derived from FBA-derived metrics are shown in Fig. 3A–F. The inter-group t values of FA were positively correlated with those of FD (r = 0.82, FDR-p < 0.001) and FDC (r = 0.66, FDR-p < 0.001). Notably, FD, log (FC), and FDC showed significant inter-group differences in a larger number of fibre bundles compared to FA. The inter-group t values of MD were negatively correlated with those of FD (r = −0.59, FDR-p < 0.001) and FDC (r = −0.41, FDR-p = 0.001). In contrast, no significant correlations were found between t values of FA and log (FC) (r = –0.01, FDR-p = 0.907), or between t values of MD and log (FC) (r = 0.12, FDR-p = 0.389). Moreover, FD and FDC showed significant inter-group differences across more fibre bundles than MD, and these fibre bundles contained the majority of those showing significant differences in FA.

Fig. 3. The correlations of t values between the tensor-based metrics and FBA-derived metrics.

In panel A–F, each data point represents the t value from the group comparison (patients vs. controls) for a single white matter tract.

Correlations between white matter metrics and clinical characteristics in FES

No statistically significant correlations were found between WM metrics and clinical characteristics, including DUP, PANSS scores, and cognitive deficits, in FES after FDR correction.

Discussion

In this study, we found that treatment-naïve individuals with FES exhibit widespread WM microstructural abnormalities across the majority of WM tracts, as assessed using the FBA framework. The observation that FBA detected a larger number of significant bundles can be explained by its inherent capacity to model multiple fibre orientations within a voxel. This makes the framework particularly sensitive to microstructural alterations in regions with crossing-fibre architecture, which are prevalent throughout the brain. Furthermore, a set of overlapping WM tracts showed statistically significant alterations in both tensor-based and FBA-derived metrics, indicating convergence of the two methods in detecting WM abnormalities. In addition, we observed statistically significant correlations of inter-group differences between FBA-derived metrics and those obtained from the conventional tensor model, highlighting underlying biological relevance of both approaches. Importantly, no significant associations were found between FBA-derived metrics (FD, FC, and FDC) and clinical characteristics, including symptom severity, DUP, or cognitive deficits, after correction for multiple comparisons. This lack of correlation implies that the observed WM abnormalities may reflect trait-like, rather than state-dependent neuropathological features. Collectively, our findings indicate that substantial WM disruption is already present at the early, untreated stage of FES and underscores the potential of the FBA framework to improve our understanding of the complex nature of WM microstructural alterations in SZ.

Among the WM tracts exhibiting significant inter-group differences, 25 fibre tracts, including right ICP, left AF, left SLF_II, CC_1 and 5, as well as bilateral CG, CST, POPT, SCP, SLF_ I, SLF_III, ST_PAR, ST_POSTC, T_POSTC, and UF showed significant reductions in FBA-derived metrics (FD, log (FC), or FDC), but no alterations in tensor-based metrics. WM constitutes approximately half of the brain’s volume and plays a crucial role in transmitting neural signals across different cortical and subcortical regions [39, 40]. These 25 fibre tracts mediate communication among the frontal, parietal, temporal lobes, brainstem, cerebellum, spinal cord, limbic system, striatum, and pontine nuclei [41–48]. Specifically, the AF and the SLF form the superior longitudinal system (SLS), which is critically involved in language function [49, 50]. The CST, SCP, and ICP support the sensorimotor system, including the control of voluntary movements and posture [51, 52]. The CG and UF are key components of the limbic system, implicated in memory processing and emotional regulation [46, 53, 54]. The CC, as the largest WM tract connecting bilateral hemispheres, underlying interhemispheric integration related to perception, consciousness, self-awareness, creativity, and higher-order cognitive functions such as attention, memory, language, and reasoning [55–57]. The remaining tracts including ST_PAR, ST_POSTC, T_POSTC, and POPT are involved in perceptual processes including visual, auditory, somatic, and spatial sense, as well as motion perception [58]. All of the aforementioned tracts exhibited significant reductions in FD, log (FC), or FDC metrics, suggesting potential deficits of brain functions associated with these tracts. Among the three FBA-derived metrics, FDC demonstrated significant inter-group differences in the largest number of fibre tracts, indicating its superior sensitivity in detecting microstructural abnormalities. This finding aligns with previous study that FDC provides a more comprehensive and sensitive assessment of fixel-wise effect, capturing both microstructural (FD) and macrostructural (FC) changes [15].

A total of 30 WM tracts demonstrated statistically significant alterations in both tensor-based metrics and FBA-derived metrics. These included the right AF, right SLF_II, left MLF, left ST_PREM, left T_PREF, left T_PREM, CC and its 3 segmentations (CC_ 2, 6, and 7), as well as bilateral ATR, FX, IFO, ILF, OR, ST_FO, ST_OCC, ST_PREF, T_OCC, and T_PAR. Among these tracts, 24 tracts including the right AF, right SLF_II, left MLF, left ST_PREF, left T_PREF, left OR, left T_PREM, left T_OCC, CC and its 3 segmentations (CC_ 2, 6, and 7), as well as bilateral ATR, IFO, ILF, ST_FO, ST_OCC, and T_PAR exhibited significant alterations in MD, and either FD or log (FC), but not in FA. FA reflects the degree of directional dependence of water diffusion within tissue and is most sensitive to changes in the primary direction of fibre organization [59]. Differences in FA typically imply alterations in the orientation-dependent characteristics of tissue microstructure [60]. In contrast, MD quantifies the average diffusivity along three orthogonal axes and reflects overall water diffusion in tissue [59]. The presence of abnormalities in MD and FBA-derived metrics (FD or log (FC)) in the absence of FA changes suggests that axonal injury may be diffusely distributed across multiple directions, without a predominant distribution along the primary fibre orientation. This diffuse pattern could result in detectable microstructural changes in MD and FBA metrics, while leaving FA unchanged. Additionally, 4 tracts including the right OR, right T_OCC, and bilateral FX showed decreased FA alongside increased MD, a pattern indicative of acute axonal injury, such as demyelination, cell death, or inflammation processes [59, 61]. In the remaining tracts, specifically the right ST_PREF and left ST_PREM, increased MD was accompanied by decreased FDC, without significant changes in FD or log (FC). This pattern may indicate combined intra-axonal and myelin loss, with damage affecting both microstructural and macrostructural components, rather than occurring in isolation.

We identified 6 WM tracts that showed statistically significant alterations exclusively in the tensor-based model, while no corresponding differences detected using the FBA framework. These tracts included the right ST_PREM, right T_PREF, right T_PREM, left ST_PREC, left T_PREC, and CC_3. It is noteworthy that DTI-derived metrics such as FA or MD are inherently non-specific and can be influenced by a wide range of biological and technical factors. These factors include axon packing density, axon caliber, microglia, inflammation, crossing fibre conditions and partial volume effects [59–62]. As a result, the presence of significant changes in FA or MD alone may not necessarily reflect isolated microstructural damage but could instead represent a complex interaction of multiple underlying processes that are not fully disentangled by the DTI model. In addition, widespread FA reductions have been well-duplicated as a robust characteristic of SZ in previous large-scale studies [63, 64]. In the present study, approximately 50% of the examined tracts showed nominal FA increases. However, the majority of these did not reach statistical significance and showed small effect sizes (Cohen’s d < 0.2 for most tracts). These subtle, non-significant trends are unlikely to represent robust neurobiological alterations. Notably, focal FA increases are not entirely unprecedented—Kelly et al. also reported increased FA in the posterior limb of the internal capsule (PLIC) in SZ patients, which they interpreted as a potential result of aberrant axonal pruning [64]. Heterogeneity in findings across studies may be partly attributed to differences in medication and illness duration. Notably, our study included drug-naïve FES patients, offering insights into WM pathology at the early stage of the disease.

The strength of the correlations between DTI and FBA metrics provides insight into the extent to which voxel-level DTI abnormalities reflect underlying fixel-level changes. The strong positive correlation between FA and FD t values (r = 0.82) and moderate-to-strong correlation between FA and FDC t values (r = 0.66) indicate alignment between these metrics, suggesting that FA abnormalities may substantially reflect fibre density deficits and the combined effects of fibre density and cross-section. However, the dissociation observed in tracts where FA showed positive t values while FD and FDC remained negative highlights that FA alone cannot disambiguate complex fibre configurations—a limitation that FBA addresses through its ability to resolve crossing fibres. The moderate negative correlations between MD and FD t values (r = −0.59), as well as between MD and FDC t values (r = −0.41), suggest that MD may capture additional information beyond fibre-specific changes, likely reflecting broader tissue properties such as inflammation or extracellular space alterations.

Finally, the specificity of FBA in capturing microstructural abnormalities come at the cost of increased methodological complexity. Unlike DTI’s relatively simple voxel-based pipeline, FBA involves multiple sophisticated steps, including sophisticated FOD modeling, fixel segmentation and reorientation, and connectivity-based smoothing. These steps require careful quality control and parameter optimization. These technical demands should be carefully weighed against research objectives when designing future studies.

Several limitations should be considered when interpreting the findings of this study. First, the cross-sectional design precludes any causal inference between WM alterations and SZ. Longitudinal studies are needed to determine whether the observed WM abnormalities are precursors, consequences, or correlates of the illness trajectory. Second, although the FBA framework offers key advantages in detecting fibre-specific microstructural and macrostructural changes, it is not without limitations. The accuracy of FBA results depends heavily on precise modeling and localization of brain anatomy. Anatomical inaccuracies or inter-individual variability in WM architecture can potentially lead to misleading interpretations [16]. Thirdly, it is important to acknowledge that DTI and FBA measure fundamentally different biological substrates—DTI reflects voxel-level water diffusion properties, while FBA provides fibre-specific metrics at the fixel level. This fundamental difference in measurement units (voxel vs. fixel), together with their distinct optimal preprocessing pipelines, renders the two approaches not directly comparable and precludes any quantitative comparisons between tensor-based and fixel-based metrics. Finally, as noted in the literature [16], FBA is ideally performed on multi-shell or high b-value (≥3000 s/mm²) data to ensure the specificity of FD. Although FBA is technically compatible with single-shell data (b = 1000 s/mm²) using SS3T-CSD as employed in this study, the use of a lower b-value may reduce FD sensitivity and interpretability due to potential contributions from extra-cellular signals. Therefore, our FD findings should be interpreted with caution.

Conclusion

In conclusion, this study employed two distinct WM analysis methods, conventional DTI and FBA, to investigate WM microstructural abnormalities in SZ patients. While both approaches revealed significant alterations, the FBA uncovered a more extensive and nuanced pattern of WM abnormalities. These findings provide compelling evidence for the enhanced specificity of the FBA framework in detecting fibre-specific pathology. By leveraging the complementary strengths of FBA and conventional tensor-based model, this study provides a more comprehensive and insightful characterization of the complex microstructural architecture of the brain in patients with SZ.

Author contributions

Substantial contributions to the conception or design of the work: ML, QW, LZ, HY, XL, WD, WG, WW, and TL; Acquisition, analysis, and interpretation of data for the work: MY, HL, and LD; Drafting the manuscript: MY and HL; Revising the manuscript critically for important intellectual content: LD, ML, QW, LZ, HY, XL, WD, WG, WW, and TL; Final approval of the version to be published: MY, HL, LD, ML, QW, LZ, HY, XL, WD, WG, WW, and TL.

Funding

This work was supported by the Key Project of the National Natural Science Foundation of China (T.L., grant number 82230046), the Key Project of Agriculture and Social Development of Hangzhou Science and Technology Bureau (T.L., grant number 20241203A14), the Zhejiang Central Guiding Local Technology Development (T.L., grant number 2024ZY01010), the Construction Fund of Key Medical Disciplines of Hangzhou (T.L., grant number 2025HZGF10 and 2025HZZD14), and the Zhejiang Key Laboratory of Clinical and Basic Research for Psychiatric Diseases (T.L., grant number 2024ZY01010 and 2024E10107).

Data availability

The data used in this article are not publicly available. Reasonable requests to access the dataset and our customized codes should be directed to the corresponding authors (TL and WW).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04277-y.