Current Clinical Applications of Structural MRI in Neurological Disorders

Abstract

Structural magnetic resonance imaging (sMRI) plays a pivotal role in the evaluation of neurological disorders by providing high-resolution anatomical information. Recent advances in quantitative postprocessing techniques have expanded the utility of sMRI beyond visual assessments by enabling the detection of subtle morphological changes associated with various neurological and psychiatric conditions. This review summarizes current clinical applications of sMRI-based analysis, including brain volumetry, shape analysis, voxel-based morphometry (VBM), surface-based morphometry, source-based morphometry, and voxel-based lesion–symptom mapping (VLSM). Volumetric and shape-based analyses allow for assessments of region-specific atrophy and subregional morphological alterations, while VBM and surface-based morphometry provide complementary insights into tissue volumes and the architecture of the cortical surface. Source-based morphometry reveals network-level patterns of structural covariance, and VLSM directly correlates lesion locations with functional outcomes, particularly in stroke. It has been demonstrated that these methodologies are clinically relevant in conditions such as Alzheimer's disease, Parkinson's disease, multiple sclerosis, epilepsy, and major depressive disorder. By quantifying structural brain alterations that are not readily detectable using conventional imaging methods, these tools improve diagnostic accuracy, support prognostication, and facilitate monitoring of treatment effects. This review highlights the growing integration of sMRI postprocessing techniques into clinical neurology.

Graphical Abstract

INTRODUCTION

Magnetic resonance imaging (MRI) is a ubiquitous technique for the high-resolution spatial visualization of brain structures. T1-weighted (T1), T2-weighted (T2), and fluid-attenuated inversion-recovery (FLAIR) MRI sequences are the most commonly used for assessing brain anatomy and pathology. These sequences highlight qualitatively different aspects of tissue properties, providing complementary views of brain morphology and pathology.¹,² By utilizing the different contrast characteristics of T1, T2, and FLAIR sequences, MRI excels at delineating different brain tissues and detecting pathological changes such as tumors, inflammation, and neurodegenerative disease. These imaging modalities are indispensable for diagnosing neurological disorders, guiding surgical planning, and advancing research in brain development, aging, and psychiatric conditions. Table 1 summarizes the complementary features of T1, T2, and FLAIR sequences.

Table 1. Comparison of T1, T2, and FLAIR MRI features.

Feature	T1 sequences	T2 sequences	FLAIR sequences
Contrast mechanism	Highlights differences in T1 relaxation times using short TR and TE. Fat is bright while water and CSF appear dark.	Emphasizes differences in T2 relaxation times using long TR and TE. Water appears bright and fat is less bright.	Suppresses free fluid (e.g., CSF) signals to enhance lesion detection, using long TI to nullify fluid signals.
Anatomical detail	High spatial resolution, excellent for evaluating normal structures. White matter appears brighter than gray matter.	Gray matter appears brighter than white matter; excellent for detecting changes in tissue water content.	Tissue contrast similar to T2, but with CSF signal suppression enhancing boundary definition.
Lesion signal characteristics	Lesions typically appear dark unless they contain fat or blood. Contrast agents enhance lesion visibility.	Lesions generally appear bright; sensitive to increased water content.	Maintains bright lesions from T2 sequences while suppressing fluid, enhancing lesion visibility.
Primary uses and advantages	Ideal for assessing normal anatomy and enhancing lesions with contrast agents	Highly sensitive to pathological changes in tissue water content; useful in detecting edema and inflammation.	Standard in detecting CNS inflammatory lesions, especially in MS, due to its ability to suppress surrounding fluid signals.
Disadvantages	Without contrast agents it might miss lesions such as edema.	Normal fluids such as CSF also appear bright, which can obscure lesions.	Longer scan times and potentially lower signal-to-noise ratio than other sequences
Typical clinical applications	Brain tumors: enhances tumor margins and invasion assessment	Brain edema and inflammation: highly sensitive to changes around tumors and in inflammatory diseases	MS: essential for detecting white-matter lesions around ventricles and under the cortex

CNS, central nervous system; CSF, cerebral spinal fluid; FLAIR, fluid-attenuated inversion-recovery; MRI, magnetic resonance imaging; MS, multiple sclerosis; T1, T1-weighted; T2, T2-weighted; TE, time to echo; TI, inversion time; TR, repetition time.

T1 sequences are typically used for high-resolution structural imaging to provide clear delineation of anatomical structures. These sequences are sensitive to the longitudinal relaxation time of tissues, meaning that areas with a high water content (e.g., cerebrospinal fluid [CSF]) appear dark, with structures such as gray matter and white matter appearing brighter. T1 imaging is excellent for observing normal brain anatomy and detecting structural abnormalities such as tumors, Alzheimer's disease (AD), multiple sclerosis (MS), infarcts, and atrophy.³,⁴,⁵,⁶,⁷ In contrast, T2 images are sensitive to the transverse relaxation time of tissues, highlighting differences in water content. In T2 images, fluids such as CSF and edematous tissue appear bright, while normal gray matter and white matter appear darker. This makes T2 imaging particularly useful for detecting pathologies involving increased fluid or edema, such as inflammation, tumors, and demyelinating diseases such as MS.⁸,⁹ FLAIR sequences are a specialized type of sequence that combine the advantages of T1 and T2 imaging while suppressing the signals from free fluid. This is achieved with an inversion-recovery pulse that nullifies signals from free fluid (e.g., CSF), improving the detection of lesions near CSF spaces and in regions of edema. FLAIR sequences are especially sensitive to subtle brain lesions (including those seen in MS) and they can highlight areas of chronic infarction and gliosis.¹⁰,¹¹

Collectively, these three types of MRI sequence yield complementary information to provide a comprehensive understanding of brain structure and pathology. They facilitate the diagnosis and monitoring of a broad spectrum of neurological conditions. Additionally, these sequences serve as valuable sources for quantitative MRI-based analyses.

In this review we introduce various structural MRI (sMRI) postprocessing techniques that are applied in clinical practice, including brain volumetry, shape analysis, voxel-based morphometry (VBM), surface-based morphometry, source-based morphometry, and voxel-based lesion–symptom mapping (VLSM). Additionally, we present an overview of research into diverse neurological and psychiatric disorders where these methods of imaging analysis have been used. This review delineates the evolving role of sMRI in neurology, underscoring its critical contributions to both clinical practice and neuroscience research.

BRAIN MRI VOLUMETRY

Brain volumetry involves quantifying the volume of various brain regions, providing critical insights into structural brain changes associated with neurological and psychiatric disorders. High-resolution MRI provides detailed anatomical data, making it an invaluable tool for volumetric analysis. Volume measurements have become essential for understanding disease mechanisms, monitoring progression, and evaluating treatment responses in conditions such as AD, MS, and schizophrenia. As this field advances, the choice between manual and automated volumetric methods remains a subject of debate, with each approach offering distinct advantages and limitations.¹²,¹³,¹⁴,¹⁵,¹⁶

Manual brain volumetry requires a trained rater to meticulously delineate regions of interest (ROIs) by hand on MRI scans. An expert typically manually outlines the boundaries of specific brain structures—such as the hippocampus, amygdala, or subcortical nuclei—on individual slices or via three-dimensional (3D) reconstruction.¹⁷,¹⁸ The main advantage of manual segmentation is the precision with which complex anatomical structures can be delineated, especially in cases of atypical anatomy or pathology.¹⁸ Indeed, manual tracing is often considered the gold standard for volumetric analysis due to its accuracy in defining abnormal brain structures. It also allows a high degree of control over the segmentation process, which can be essential in research and clinical cases where accuracy is paramount. However, manual volumetry is extremely time-consuming, and its results are highly operator-dependent. It requires considerable expertise and can be subject to significant observer variability, particularly when multiple raters are involved or when region boundaries are ambiguous. The time-consuming nature of manual methods make them impractical for large datasets and in longitudinal studies where numerous scans must be processed.¹⁹

On the other hand, automated brain volumetry uses software algorithms to segment and quantify brain structure volumes without manual intervention. These methods typically rely on image registration, statistical modeling, and machine-learning techniques to identify anatomical regions and calculate their volumes.²⁰,²¹,²²,²³,²⁴ The primary advantages of automated volumetric methods are efficiency and scalability. Automated techniques can rapidly process large datasets, making them well-suited for studies involving numerous subjects as well as clinical applications that require high-throughput analysis. Moreover, automated methods reduce the risks of observer bias and interrater variability, providing consistent and reproducible results. This objectivity is particularly beneficial in multicenter studies and clinical trials where standardization across sites is crucial.²³ Automated methods are also far less labor-intensive than manual segmentation, significantly reducing both the time and cost of data processing. Numerous automated tools are available for different MRI sequences and brain regions, including software specifically designed to segment structures such as the hippocampus, cortical gray matter, and white matter.²⁵,²⁶ However, automated volumetry is not without limitations. These methods may struggle with complex and atypical brain morphologies, such as in patients with severe atrophy, congenital malformations, or lesions involving anatomical distortion. Moreover, the accuracy of automated segmentation is highly dependent on the input image quality, and variations in scan parameters or resolution can also affect the performance. Additionally, some automated algorithms still require human oversight or correction to ensure precision, especially in regions with indistinct boundaries.²⁷

Choosing between manual and automated volumetry methods ultimately depends on the specific research or clinical needs. Manual segmentation remains the preferred method for studies requiring the highest precision, particularly when brain regions are poorly defined or when significant anatomical distortion is present. In such cases, manual methods allow for nuanced, case-by-case delineation to ensure every structure is accurately measured despite the presence of irregularities. Conversely, automated volumetric methods are well-suited to large-scale studies, longitudinal research, and clinical environments where speed, reproducibility, and consistency are critical. Automated techniques can handle volumes of data that would be infeasible to process manually. Furthermore, the accuracy of automated tools is increasing as computational models continue to improve, making them more reliable even in cases with subtle abnormalities.²⁸,²⁹,³⁰

MRI-based volumetric analysis, particularly of small brain structures such as the hippocampus, is highly sensitive to methodological variability including the pixel-inclusion criteria applied during boundary delineation and the technical specifications of the analysis platform.²³ As demonstrated by Jack et al.,¹⁶ even minor differences in pixel-counting strategies can lead to substantial volume discrepancies for small structures. Furthermore, Gronenschild et al.³¹ showed that variations in FreeSurfer software versions, workstation types, and operating-system versions can significantly influence estimates of volumes and cortical thicknesses. Therefore, to ensure methodological consistency and data comparability, MRI-based volumetric analyses should be performed using the same software version, operating system, and hardware configuration throughout a particular study. Fig. 1 shows the significant variations in results that can arise from using different pixel-counting strategies or different software versions and operating systems.

Fig. 1. Variability in hippocampal and subcortical volume measurements: effects of ROI definitions and software versions. A: Schematic of hippocampal cross-sectional area measurements derived from traced ROIs using a cylindrical model. Three different pixel-inclusion methods demonstrate the considerable variability introduced by ROI boundary definitions. Adapted from Jack et al. Magn Reson Imaging 1995;13:1057-1064.¹⁶, under the terms of the Creative Commons License (CC BY NC ND). B: Comparison of estimated subcortical gray-matter volumes between FreeSurfer software versions 4.3.1 (upper row) and 5.0.0 (lower row). Adapted from Gronenschild et al. PLoS One 2012;7:e38234.³¹, under the terms of the Creative Commons License (CC BY). ROI, region of interest.

Brain volumetry based on MRI data is a cornerstone technique for studying changes in brain structure between healthy and disease states. While manual volumetry remains highly accurate, it is slow and susceptible to observer variability. Automated volumetry provides faster and more-reproducible results, but it may struggle with complex brain morphologies and therefore require careful validation. Automated methods are expected to become increasingly sophisticated as neuroimaging technologies continue to evolve, leading to the more-widespread and reliable use of volumetric analysis in both research and clinical practice.

Clinical applications of volumetry

Volumetric MRI assists early diagnosis and monitoring in neurodegenerative disorders. Hippocampal volumetry aids early AD detection by identifying characteristic atrophy patterns, and it can be used to track disease progression and evaluate treatment responses. In MS, global and regional brain atrophy serves as a marker of disease progression alongside lesion assessments. Additionally, subtle regional volume loss in early Parkinson's disease (PD) may predict cognitive decline, and volumetry can be used to objectively evaluate treatment effects in clinical trials of dementia.

Limitations of volumetry

Variability in the scanners and software used for brain volumetry results in volume estimates that are inconsistent across sites or over time. Automated segmentation errors may occur, particularly in cases with severe atrophy or anatomical distortion, while precise manual volumetry remains impractical for routine use. Volumetric methods may miss localized subregional changes and require careful normalization for intracranial volume differences. Standardization and rigorous quality control are critical to addressing these limitations.

TOTAL INTRACRANIAL CAVITY VOLUME

The total intracranial cavity volume (TICV) is used as a surrogate for an individual's premorbid brain size, which is particularly relevant when analyzing conditions involving cerebral atrophy.³²,³³ Estimating a patient's premorbid intracranial volume is crucial in volumetric and morphometric studies of diseases characterized by brain volume loss. Because the skull expands to accommodate brain growth during normal development, TICV reflects a person's peak brain volume—one that is not reduced by aging or pathological atrophy. Individuals with larger TICVs generally have larger absolute brain volumes, and so volume measurements of specific structures (e.g., hippocampus) often need to be adjusted for TICV in comparative studies.³⁴,³⁵

While T2 MRI can clearly delineate the dura mater (which closely approximates the inner boundary of the skull), high-resolution 3D T2 sequences are not commonly acquired in clinical protocols. Eritaia et al.³⁶ described an optimal method for manually measuring TICV in standard T1 MRI (Fig. 2). However, manual TICV annotation is extremely labor-intensive and time-consuming, and it also requires extensive anatomical information.

Fig. 2. Manual TICV segmentation. A 5-mm-thick sagittal MRI slice (A) reconstructed from a high-resolution volumetric T1-weighted scan was used to delineate the total intracranial cavity. The inner boundary of the skull (dura mater) was traced manually (B), and the lateral limits were defined as the most-lateral slices containing brain parenchyma. The lower tip of the cerebellum was defined as the inferior limit. To establish the inferior boundary on head-tilt-corrected sagittal images, a horizontal line was drawn across the midbrain to include the cerebellar tip (black arrow). Adapted from Tae et al. Investig Magn Reson Imaging 2009;13:63-73.³⁹, under the terms of the Creative Commons License (CC BY NC). MRI, magnetic resonance imaging; TICV, total intracranial cavity volume.

Various automated techniques have been developed in efforts to overcome the challenges of manual TICV measurements. However, there are often notable discrepancies in TICV values between those obtained manually and those obtained automatically.³⁵,³⁷,³⁸ Some linear-transform-based approaches, which do not directly segment all structures within the intracranial cavity, tend to underestimate TICV in patients with significant cerebral atrophy.³⁹

Methods for automated TICV estimation that are more accurate and reliable have recently been introduced.⁴⁰,⁴¹ These segmentation-based approaches provide TICV measurements that closely reflect the true intracranial volume.

SHAPE ANALYSIS

Shape analysis represents a significant advancement in neuroimaging, offering detailed statistical assessments of morphometric alterations in specific brain subregions. Unlike traditional volumetry, which provides overall volume measurements for a given brain structure, shape analysis employs 3D modeling to examine the precise boundaries and contours of individual structures. This makes it possible to detect localized atrophy or hypertrophy within a region, revealing subtle shape changes that are indicative of pathological processes even when the total volume is relatively preserved. As illustrated in Fig. 3, regional shape contractions of subcortical nuclei have been found in amyotrophic lateral sclerosis (ALS) patients relative to normal healthy subjects. Consequently, shape analysis provides a more-nuanced understanding of brain morphology in both healthy and diseased states, enabling statistical examinations of localized anatomical changes within larger regions.¹⁷,²⁵,⁴²,⁴³,⁴⁴,⁴⁵

Fig. 3. Three-dimensional visualizations of subcortical structures in ALS patients. Relative to healthy controls, the ALS patients exhibit regional shape contractions affecting both globus pallidi, the right putamen, and the right nucleus accumbens. Adapted from Tae et al. J Clin Neurol 2020;16:592-598.⁴⁵, under the terms of the Creative Commons License (CC BY NC). ALS, amyotrophic lateral sclerosis.

Implementing shape analysis generally involves several steps. The brain is first segmented into distinct ROIs, and then specialized algorithms extract the 3D shape of each region, capturing its complex curvature and geometric properties. These shapes can then be compared across subjects or time points to reveal differences in morphology that may reflect underlying neurobiological processes.⁴³,⁴⁶,⁴⁷

A key advantage of shape analysis is its high sensitivity to localized deformations within brain subregions, which is crucial for analyzing diseases that target specific brain areas. For example, in AD—which has a hallmark of hippocampal atrophy—shape analysis can pinpoint the precise subfields of the hippocampus that are most affected by neurodegeneration.⁴⁸,⁴⁹ Shape analysis has been widely applied in numerous neurological and psychiatric conditions, including AD,⁵⁰,⁵¹ ALS,⁴⁵ epilepsy,⁵² PD,⁵³ MS,⁵⁴ tinnitus,⁴⁴ major depressive disorder (MDD),¹⁷,⁵⁵,⁵⁶,⁵⁷ and schizophrenia.⁵⁸

While volumetric MRI techniques have greatly improved our understanding of brain structure, shape analysis offers distinct benefits. Volume measurements might sometimes not detect small but important localized changes. Also, a brain region might undergo atrophy without a substantial change in its overall volume, despite its shape being significantly altered. Shape analysis can detect these shape changes, which often serve as early indicators of pathological processes. This limitation of volumetric analysis is due to it typically examining large, predefined ROIs, potentially missing smaller but clinically significant alterations.⁵⁹,⁶⁰,⁶¹ In contrast, shape analysis excels at detecting fine-grained, localized deformations at a high spatial resolution, providing more-sensitive biomarkers of disease progression.

Shape analysis enables quantification of surface-based metrics and other detailed structural parameters that volumetric methods cannot capture. It has demonstrated high sensitivity in identifying morphological changes across various neurological and psychiatric disorders, including neurodegenerative diseases, developmental abnormalities, and traumatic brain injuries. This increased sensitivity is potentially significant for improving diagnostic accuracy, monitoring disease progression, and developing targeted therapeutic interventions.¹⁷,⁴⁴,⁴⁵

Clinical applications of shape analysis

The high sensitivity of shape analysis in detecting localized structural changes aids early diagnoses and prognoses. In AD, shape analysis can identify subfield-specific hippocampal atrophy that distinguishes AD from normal aging and predicts progression from mild cognitive impairment (MCI). It can also reveal subtle hippocampal deformations in depression, potential biomarkers in epilepsy when MRI produces normal findings, and patterns of subcortical atrophy in PD and ALS, supporting disease staging and subtype characterization.

Limitations of shape analysis

Shape analysis depends on a precise initial segmentation process, with inaccuracies therein leading to misleading results. The lack of standardization across algorithms complicates comparisons between studies. Additionally, interpreting shape changes when there is no overall volume change is challenging for clinicians. Shape analysis is typically applied at the group level, since its use for individualized diagnoses remains restricted by normal anatomical variability between subjects. Finally, the computational complexity of shape analysis and its specialized software requirements hinder its widespread clinical implementation, emphasizing the need for user-friendly pipelines and standardized protocols.

VOXEL-BASED MORPHOMETRY

VBM is a widely used neuroimaging technique for analyzing sMRI data to compare local proportions or volumes of gray matter, white matter, or CSF across the brain. Although the VBM framework was formally proposed by Ashburner and Friston⁶² in 2000 and Good et al.,⁶³ earlier work by researchers such as Andreasen et al.⁶⁴ laid the foundation by developing automated techniques for sMRI-based analysis. For example, Andreasen's team created an „averaged brain… in a standardized stereotaxic space⁶⁵ to compare brain structures between healthy individuals and patients with schizophrenia—an approach conceptually similar to VBM. This evolution from early manual and semiautomated methods to the fully automated VBM pipelines used today highlights the progression of neuroimaging analysis techniques.

The VBM process involves several key steps:⁶²,⁶³

• 1) Spatial normalization in which high-resolution T1 images are aligned to a standard brain template, placing all scans into the same stereotaxic space.

  2) Segmenting normalized images into gray matter and white matter (and sometimes CSF) using algorithms that classify tissue types.

  3) Smoothing the segmented images with a Gaussian kernel to reduce noise and account for anatomical variability between individuals, facilitating voxel-wise statistical analysis.

  4) Voxel-wise statistical analyses to compare gray-matter concentrations between groups or to correlate with covariates, with corrections for multiple comparisons typically applied using Gaussian random field theory.

In addition, techniques such as threshold-free cluster enhancement can be applied to improve cluster detection without relying on a hard voxel-wise threshold,⁴¹,⁶⁶ and nonparametric permutation tests can be used to provide robustness when the assumption of a normal distribution is not valid.⁶⁷

VBM has been extensively applied in various neurological and psychiatric conditions. For example, it has been used to investigate structural brain changes in epilepsy,⁵²,⁶⁰ obstructive sleep apnea,⁶⁸ AD,⁶⁹,⁷⁰,⁷¹ MS,⁷² and PD.⁷³,⁷⁴ Fig. 4 illustrates a decreased white-matter concentration and shape contractions of the corpus callosum in patients with MDD. VBM has also identified patterns of gray-matter atrophy in MCI,⁷⁵ such as significant volume losses in the entorhinal cortex and frontal lobes. In healthy individuals, VBM techniques have been employed to study the effects on brain structure of intensive practicing by athletes such as basketball players (Fig. 5).⁷⁶

Fig. 4. VBM of gray and white matter in MDD. A: VBM analysis of patients with MDD showed reduced white-matter concentrations in the genu and splenium of the corpus callosum. B: Shape analysis of the same patients revealed localized shape contractions in the rostrum and splenium of the corpus callosum. C: VBM also demonstrated decreases in the gray-matter concentration in multiple regions, including the medial prefrontal cortex, lateral prefrontal cortex, orbitofrontal cortex, anterior cingulate gyrus, insular cortex, lateral temporal cortex, and mesial temporal lobe. Adapted from Lee et al. Psychiatry Investig 2020;17:941-950.⁵⁷, under the terms of the Creative Commons License (CC BY NC). MDD, major depressive disorder; VBM, voxel-based morphometry.

Fig. 5. VBM in basketball players. A: The cortex was thicker in basketball players than controls, notably in both pericentral gyri and the paracentral lobules. B: The FD was also higher in basketball players, with a significantly increased FD observed in both precentral gyri. The color bar represents t values. Adapted from Kim et al. J Korean Med Sci 2022;37:e86.⁷⁶, under the terms of the Creative Commons License (CC BY NC). FD, fractal dimension.

A notable limitation of VBM is the introduction of substantial variability by using different image-processing pipelines, which interferes with reproducibility. For example, one study found considerable differences in gray-matter findings when the same dataset was processed using four common VBM pipelines including CAT,¹²,⁴¹ FSL-VBM,⁷⁷ and sMRIPrep.⁷⁸ To address these challenges, it has been suggested that VBM be integrated with other methods such as surface-based morphometry to provide a more-comprehensive understanding of brain structure.⁷⁹

A commentary article has provided practical guidelines for conducting and presenting VBM studies to ensure transparency, reproducibility, and utility.⁸⁰ Those authors propose the following 10 key rules for VBM researchers:

• 1) Clearly state the study's rationale and describe the data.

  2) Explain how brain segmentations are produced.

  3) Describe the method used for intersubject spatial normalization.

  4) Make the statistical analyses transparent.

  5) Clearly state the method used to correct for multiple testing.

  6) Present results unambiguously.

  7) Justify any nonstandard analyses.

  8) Guard against common pitfalls.

  9) Report any exclusions, and justify them.

  10) Interpret results cautiously while considering potential biases and limitations.

VBM is a powerful tool for assessing brain structure that has been widely adopted in both research and clinical settings. Nevertheless, its limitations—particularly the variability associated with different analysis pipelines—should be addressed through systematic validation and methodological improvements. Combining VBM with other morphometric techniques may further improve our understanding of brain structure and its relationship to neurological and psychiatric conditions.⁸¹

SOURCE-BASED MORPHOMETRY

Source-based morphometry is a multivariate statistical approach used in sMRI to identify patterns of covariance among brain regions. It builds upon the principles of VBM but uses independent-components analysis to decompose the data, yielding a network-level view of structural variations between individuals.⁸²,⁸³ In practice, source-based morphometry separates MRI data into spatially independent components that represent groups of voxels whose gray-matter volume or density covaries across subjects. This approach allows researchers to explore structural networks that might be associated with specific cognitive functions or disease processes. Source-based morphometry has been particularly valuable in studies of complex disorders such as AD, PD, MS, and schizophrenia, revealing linked differences between gray matter and white matter that traditional univariate analyses might miss.⁸⁴ Fig. 6 is an example of source-based morphometry in patients with PD, showing a decrease in the white-matter concentration with advancing age (unpublished data, Tae, 2018).

Fig. 6. In patients with Parkinson's disease, one gray-matter-independent component was positively correlated with age, indicating a decrease in the white-matter concentration with advancing age. The color bar represents Z values (unpublished data, Tae, 2018).

While both source-based morphometry and VBM analyze brain morphology, the former's multivariate approach offers certain advantages. VBM examines each voxel independently, potentially missing interrelationships between distant brain regions. In contrast, source-based morphometry captures covariance among voxels across the entire brain, thereby providing insight into network-level structural patterns—insights that are especially relevant for complex neurological and psychiatric conditions.

Recent methodological advances have further improved the utility of source-based morphometry. For example, constrained source-based morphometry incorporates prior anatomical information to guide the component-extraction process, combining the strengths of ROI and data-driven approaches.⁸⁵ Additionally, distributed implementations such as decentralized constrained source-based morphometry enable collaborative analyses of large-scale datasets while preserving data privacy. As computational methods and machine-learning techniques continue to evolve,⁸⁶ source-based morphometry is expected to become more sophisticated and more widely used in both research and clinical settings. Future developments may further improve its diagnostic accuracy and provide deeper insights into the neurobiological bases of neurological disorders.

Overall, source-based morphometry represents a pivotal development in neuroimaging by offering a network-level understanding of structural brain alterations. This perspective is fundamental for advancing our understanding of complex brain disorders and could ultimately contribute to improved diagnoses and treatments.

Clinical applications of source-based morphometry

Source-based morphometry offers insights into network-level brain changes relevant to diseases with widespread pathology. In AD, it identifies coatrophy patterns across frontal, parietal, and temporal regions that are linked to cognitive decline. In schizophrenia, source-based morphometry reveals gray-matter networks associated with symptoms and connectivity disruptions. PD and MS studies have revealed structural covariance involving the basal ganglia and cortex that align with clinical subtypes. While not yet applied routinely in clinical practice, source-based morphometry holds promise in identifying system-level biomarkers to aid diagnoses and patient stratification.

Limitations of source-based morphometry

Source-based morphometry remains largely restricted to research settings due to several challenges. Interpreting the independent components is difficult since they often span anatomically diverse regions and require specific expertise to correlate them with clinical data. Results can vary depending on parameters such as the number of extracted components, affecting reproducibility. Reliable analyses typically demand large samples, restricting the use of source-based morphometry in smaller studies. Moreover, source-based morphometry tools are not widely integrated into clinical workflows, and clinicians might not be familiar with interpreting network-level patterns. Standardization, improved usability, and validation against clinical outcomes are essential for advancing the clinical relevance of source-based morphometry.

SURFACE-BASED MORPHOMETRY

Surface-based morphometry is a powerful neuroimaging technique for analyzing the structural characteristics of the cerebral cortex. It quantifies various cortical parameters such as the cortical thickness, sulcal depth, sulcal width, fractal dimension (FD), and gyrification index, each of which is important for understanding disease-related changes in cortical morphology.⁴¹,⁸⁷,⁸⁸,⁸⁹ Surface-based morphometry has been applied in conditions such as epilepsy, fibromyalgia, Meige syndrome, and AD to investigate alterations in cortical structures.⁹⁰,⁹¹,⁹²,⁹³

Cortical thickness

The cortical thickness is the distance between the white-matter surface and the pial surface of the cortex, which is a critical parameter in many neurodegenerative and neurodevelopmental conditions (Fig. 5A). For example, children with benign epilepsy with centrotemporal spikes (BECTS) show extensive cortical thinning in the bilateral frontal and temporal lobes, which correlates with deficits in language and memory.⁹⁴ Patients with fibromyalgia exhibit cortical thinning in the right primary motor cortex, which is negatively correlated with pain severity and disease duration.⁹⁵ In AD, pronounced cortical thinning in the frontal, temporal, and parietal lobes has been associated with impairments in verbal memory and executive function.⁹⁶

Sulcal depth and width

The sulcal depth is the distance from the cortical surface to the deepest point of a sulcus, whereas the sulcal width is the distance between the opposite sides of a sulcus. Disease-related changes in these measures have been observed. Patients with BECTS have an increased sulcal depth in the left fusiform gyrus, indicating possible neurodevelopmental disturbances.⁹⁴ In fibromyalgia, an increased sulcal depth has been noted in the right precuneus.⁹⁵ Studies of AD have highlighted changes in the sulcal depth in the temporal and supramarginal gyri that are correlated with language and executive dysfunction.⁹⁶

Fractal dimension

The FD of the cortex is a measure of its geometric complexity (Fig. 5B).⁹⁷ In autism spectrum disorder comorbid with attention deficit hyperactivity disorder, an increased FD has been found in the right fusiform gyrus, indicating neurodevelopmental features that are distinct from autism alone.⁹⁸ Patients with Meige syndrome also show abnormal cortical complexity, which may contribute to the motor and nonmotor symptoms of the disorder.⁹²

Gyrification index

The gyrification index quantifies the degree of cortical folding, and changes in gyrification have also been reported in disease states. BECTS patients exhibit increased gyrification in certain regions, which is negatively correlated with verbal IQ.⁹⁴ Patients with probable idiopathic rapid-eye-movement sleep behavior disorder show reduced gyrification across several cortical areas; these changes could serve as a prognostic biomarker for neurodegeneration.⁹⁹ Moreover, patients with MDD who underwent electroconvulsive therapy developed increased cortical folding (hypergyrification) in the left middle temporal gyrus, suggesting that cortical neuroplasticity had increased after the treatment.¹⁰⁰

Surface-based morphometry provides invaluable insights into cortical morphology, but it is important to consider potential limitations and sources of bias of this approach. Factors such as sample size, image acquisition protocols, and statistical methods can influence the obtained results, and must be accounted for when interpreting the findings. Moreover, combining VBM with surface-based morphometry has been proposed for obtaining complementary information on gray-matter volume and cortical structures,¹⁰¹ potentially improving the detection of morphological changes and leading to a better overall understanding of neurological diseases.

Clinical applications of surface-based morphometry

The cortical thickness is a key surface-based morphometry metric that could be a useful biomarker in neurodegenerative and movement disorders. In PD, thinning in frontal and parietal regions correlates with the disease stage and cognitive status, aiding prognoses and subtype differentiation. Regional thinning—especially in the entorhinal cortex—supports early diagnoses of AD and MCI. Surface-based morphometry has also shown promise in conditions such as fibromyalgia, where cortical thinning in pain-related regions reflects the symptom severity. These findings highlight the potential of surface-based morphometry to provide spatially detailed, disease-relevant insights in clinical practice.

Limitations of surface-based morphometry

Surface-based morphometry faces challenges similar to volumetry, with additional issues that are specific to cortical measures. Estimates of the cortical thickness can vary between software platforms and will be affected by the scan quality and motion artifacts, requiring strict quality control. Interpreting cross-sectional data is difficult due to natural anatomical variability, and performing diagnoses in individual patients is restricted by the absence of normative databases. Cortical thinning is only an indirect marker of pathology and can be influenced by nonneuronal factors. Longitudinal studies are also vulnerable to scanner variability and measurement noise. Standardized protocols and robust reference datasets are needed to improve the clinical utility of surface-based morphometry.

VOXEL-BASED LESION–SYMPTOM MAPPING

VLSM is an advanced approach for correlating focal brain lesions with clinical deficits.¹⁰²,¹⁰³ This method statistically assesses how the presence or absence of a lesion in each specific voxel relates to behavioral or clinical outcomes. In practical applications, patients with a lesion involving a specific voxel in the brain are compared with patients without a lesion at that voxel to determine whether the two groups differ in a given behavioral score. The resulting statistical map highlights brain regions where damage is associated with a significant impact on the function of interest. VLSM can thereby pinpoint the critical anatomical structures necessary for certain functions (e.g., motor control, balance, or gait after stroke) and can predict recovery outcomes such as the potential for poststroke language rehabilitation.¹⁰⁴,¹⁰⁵,¹⁰⁶,¹⁰⁷,¹⁰⁸

Performing VLSM typically involves several steps. First, each patient's lesion is delineated (either manually or with automated tools) to create a binary lesion map.¹⁰⁹ Second, the lesion data are compiled on a common voxel grid in which each voxel is labeled as either lesioned or intact.¹¹⁰ Third, all lesion maps are spatially normalized to a standard brain template so that comparisons can be made across subjects within the same region.¹¹¹

Researchers have also begun integrating VLSM with other imaging modalities to improve our understanding of brain function and recovery. For example, combining VLSM with functional MRI¹¹² or diffusion-tensor imaging¹¹³,¹¹⁴ can provide a more-comprehensive view of how structural damage, functional activity, and connectivity are related to clinical outcomes (Fig. 7). This multimodal approach leverages the strengths of different techniques to give a more-complete picture of brain plasticity and recovery after injury or disease.

Fig. 7. Spatial relationship between a lesion and the corticospinal tract in patients with CRPS. The image shows a subtracted percentage map illustrating differences in lesion overlap between CRPS patients and controls, overlaid on the corticospinal tract (in green). The color bar indicates the degree of lesion overlap. Images are shown according to neurological conventions, demonstrating significant overlap of the lesion site with the descending corticospinal tract pathways. Adapted from Lee et al. Sci Rep 2021;11:13093.¹¹³, under the terms of the Creative Commons License (CC BY). CRPS, complex regional pain syndrome.

VLSM has proven to be a powerful tool for investigating brain–behavior relationships in patients with focal brain damage. Ongoing advancements in computational techniques will lead to refinements in VLSM that are expected to further improve its sensitivity and specificity in identifying the brain regions that are critical for specific cognitive and motor functions. Future developments may include improved spatial normalization methods, more-robust statistical approaches (e.g., better handling of multiple comparisons), and deeper integration of VLSM with other neuroimaging modalities—all aimed at providing a more-comprehensive understanding of how brain lesions lead to clinical symptoms.

Clinical applications of VLSM

VLSM helps to identify the brain regions that are critical for specific functions, aiding prognoses and rehabilitation planning, especially after stroke. For example, lesions in left perisylvian areas are linked to persistent aphasia, which can guide speech therapy. Similarly, damage to the internal capsule or motor cortex predicts the potential for motor recovery. Beyond stroke, VLSM has revealed lesion patterns associated with poststroke pain syndromes such as complex regional pain syndrome, particularly when this involves the corticospinal tract. By mapping lesion–deficit relationships in a voxel-wise manner, VLSM supports individualized rehabilitation strategies and outcome predictions.

Limitations of VLSM

VLSM faces key challenges, especially regarding spatial normalization in brains with large or irregular lesions, which can reduce the mapping accuracy in the presence of lesion-related anatomical distortion. The results are sensitive to the registration algorithms utilized, and require large and diverse samples to ensure adequate voxel coverage. The need to correct for multiple comparisons may reduce the sensitivity, and controlling for lesion size is essential to avoid confounding effects. The clinical application of VLSM to specific cases is restricted by its reliance on group-level data. Improvements in normalization methods and integration with connectome data may improve its precision and clinical relevance.

EMERGING TRENDS AND FUTURE DIRECTIONS

Several emerging trends have recently begun to shape the future of sMRI-based analysis in clinical practice. These include applying deep learning to image analysis, combining sMRI with other modalities (e.g., multimodal imaging and biomarkers), and developing explainable artificial intelligence (XAI) techniques to interpret the output from automated analyses.

Deep learning for sMRI-based analysis

Recent advances in deep learning, particularly in convolutional neural networks (CNNs), have significantly improved sMRI-based analyses. CNNs can accurately identify complex brain patterns, improving tasks such as predicting the brain age, disease classification, and lesion segmentation. Notably, CNN-based models demonstrated superior accuracy in distinguishing AD from healthy controls using sMRI alone, even predicting amyloid pathology, which could support early noninvasive screening.¹¹⁵ Such methods also show promise in diagnosing brain tumors and autism spectrum disorder, and predicting outcomes after brain injury. There are ongoing efforts focused on validating these tools across diverse populations and scanners, highlighting their potential as rapid automated clinical-decision-support systems.

Multimodal imaging integration

Integrating sMRI with additional biomarkers and imaging modalities improves its diagnostic and prognostic capabilities, particularly in neurodegenerative diseases. For example, combining the MRI-derived hippocampal volume with positron-emission tomography (PET) amyloid or tau biomarkers substantially improves AD diagnoses and predictions of cognitive decline.¹¹⁶ Similar multimodal strategies have been explored in MS, traumatic brain injury, epilepsy, and schizophrenia, incorporating MRI with CSF biomarkers, PET, electrophysiology, and genetics. The rise of hybrid PET–MRI scanners facilitates simultaneous molecular and structural imaging, which notably improves lesion characterization in oncology and movement disorders. It is likely that such multimodal frameworks will dominate future clinical workflows, providing comprehensive diagnostic and prognostic insights.

XAI and interpretability

As complex artificial intelligence (AI) models become more common in sMRI-based analyses, ensuring interpretability has emerged as a critical priority, leading to the development of XAI.¹¹⁷ XAI techniques such as saliency maps illustrate how AI-derived decisions (e.g., identifying AD) correlate with known pathological regions, thereby enhancing the trust in clinicians and the understandability of their explanations. By converting complex AI results into comprehensible visual annotations or human-interpretable explanations, XAI supports clinicians in making informed diagnoses. Future clinical implementations of AI will probably require built-in explainability, which will facilitate transparent, reliable, and clinically actionable interpretations of sMRI findings.

CONCLUSION

sMRI postprocessing techniques—including brain volumetry, shape analysis, VBM, surface-based morphometry, source-based morphometry, and VLSM—have become indispensable tools in clinical neurology (Table 2). By quantifying subtle brain changes that are not apparent on conventional scans, these methods can improve the accuracy of diagnoses and prognoses, and enable the objective monitoring of disease progression and treatment responses. Volumetric and morphometric analyses help to detect characteristic atrophy patterns in neurodegenerative diseases such as volume loss, region-specific shape contractions, and cortical thinning, which support early diagnosis and guide clinical decision-making. In acute lesions such as stroke, sMRI mapping—including VLSM—localizes damage and correlates it with clinical deficits, aiding immediate diagnoses, predicting outcomes, and informing interventions such as targeted rehabilitation. Quantitative MRI measures have even identified subtle structural alterations associated with the illness burden in psychiatric conditions such as MDD, providing objective biomarkers to complement clinical assessments and track treatment effects. These quantitative sMRI techniques now form an essential component of the clinical toolkit for neurological disorders. They improve the ability of clinicians to make informed diagnoses, anticipate disease prognoses, and evaluate therapeutic efficacy across a broad range of conditions.

Table 2. Summary of sMRI techniques.

sMRI technique	Core methodology	Major clinical applications	Key strengths and limitations	Software	References	Recommended use cases
Brain volumetry	Quantifies the volume of brain regions using high-resolution MRI. Manual methods are precise but time-consuming; automated methods are scalable but may struggle with atypical anatomy.	AD, MS, schizophrenia, and major depressive disorder. Used to detect structural atrophy and monitor progression.	Provides objective volume measurements for disease biomarkers. Manual methods are accurate but labor-intensive; automated methods are efficient but may require manual correction.	FreeSurfer, SPM (CAT12), FSL, and volBrain; ITK-SNAP for manual segmentation	Jack et al.15 1992; Amland et al.,20 2024; Lee et al.,27 2021	Tracking atrophy in dementia and MS; volumetric biomarkers in clinical trials or large-scale datasets.
Shape analysis	Analyzes the 3D shape of brainstructures to detect localizedatrophy or hypertrophy, often more sensitive to regional subtle changes than volume alone.	AD, ALS, epilepsy, Parkinson’s disease, depression, and schizophrenia. Used to assess shape alterations in subcortical and hippocampal regions.	Highly sensitive to localized deformations even when there is no total volume change. Restricted to specific structures; requires segmentation and specific xpertise.	FreeSurfer, FSL, ENIGMA-SHAPE, and SPHARM-PDM	Scher et al.,48 2007; Gutman et al.,58 2022; Tae et al.,17 2011; Tae et al.,45 2020	Detecting localized morphological changes in neurodegenerative and psychiatric disorders.
Voxel-based morphometry	Whole-brain voxel-wise comparison of gray- or white-matter concentrations or volumes normalized to a standard template, often used for group comparisons and correlations with clinical variables.	AD, Parkinson’s disease, sleep apnea, and epilepsy. Used for detecting regional gray-matter atrophy and group-level statistical differences.	No need for ROI definition; suitable for exploratory analyses. Results depend on preprocessing and statistical corrections; may blur focal changes.	SPM, CAT12, and FSL-VBM; MRIcron for visualization	Ashburner and Friston,62 2000; Zhang et al.,69 2019; Good et al.,63 2001	Exploratory group analyses for identifying patterns of atrophy linked to disease or behavior.
Surface-based morphometry	Quantifies cortical metrics such as thickness, surface area, sulcal depth and width, fractal dimension, and gyrification index using surface-based modeling of the cortex.	AD, fibromyalgia, Meige syndrome, epilepsy, and ADHD. Used to detect fine cortical alterations associated with cognition and pain.	Sensitive to fine-grained cortical differences; good for surface-level disorders. Computationally intensive and affected by segmentation quality.	FreeSurfer, CIVET, and BrainSuite; SurfStat for group statistics	Fischl and Dale,87 2000; Ossenkoppele et al.,93 2019; Tu et al.,95 2022	Cortical studies of neurodevelopment, dementia, pain, and psychiatric disorders.
Source-based morphometry	Uses ICA to extract covarying structural patterns across subjects, enabling network-level morphological assessments.	AD, Parkinson’s disease, MS, and schizophrenia. Used to identify structural network abnormalities.	Captures multiregional patterns missed by univariate methods; provides network-level insights. Interpretation of components can be challenging; data intensive.	GIFT toolbox	Xu et al.,82 2009; Gupta et al.,83 2019; Saha et al.,85 2023	Mapping brain-wide structural networks in complex diseases such as schizophrenia and AD.
Voxel-based lesion–symptom mapping	Statistically compares behavioral scores based on lesion presence at specific voxels across patients, identifying brain regions critical for specific functions.	Stroke, aphasia, and complex regional pain syndrome. Used to correlate lesion locations with motor, language, and sensory deficits.	Links lesion location to function using group data; powerful for mapping functional neuroanatomy. Needs large samples; restricted to lesion-based populations.	MRIcron, NiiStat, and ITK-SNAP for lesion masks; voxel-wise nonparametric statistics	Bates et al.,102 2003; Geva et al.,106 2012; Gleichgerrcht et al.,114 2017	Identifying critical regions for recovery of language or motor function poststroke; guiding rehabilitation.

3D, three-dimensional; AD, Alzheimer's disease; ADHD, attention-deficit/hyperactivity disorder; ALS, amyotrophic lateral sclerosis; ICA, independent-components analysis; MRI, magnetic resonance imaging; MS, multiple sclerosis; ROI, region of interest; sMRI, structural MRI.

Availability of Data and Material

Data sharing not applicable to this article as no datasets were generated or analyzed during the study.