Quantifying individualized deviations of brain structure in patients with multiple neurological diseases from normative references

doi:10.21203/rs.3.rs-5064332/v1

Download PDF

Article

Quantifying individualized deviations of brain structure in patients with multiple neurological diseases from normative references

https://doi.org/10.21203/rs.3.rs-5064332/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Quantifying individualized deviations in the brain structure of patients with brain disorders from those of normal individuals is crucial for understanding disease pathology and guiding personalized management. In this study, we aimed to establish Chinese-specific normative references using 3D T1-weighted magnetic resonance images of 12,060 healthy controls (HCs) and to quantify the deviations in brain structure of 3,245 patients with multiple neurological diseases, including neurodegenerative diseases (mild cognitive impairment [n = 212], Alzheimer's disease [n = 467], and Parkinson's disease [PD, n = 1,263]), cerebrovascular disease (cerebral small vessel disease [n = 498]), and neuroinflammatory diseases (multiple sclerosis [MS, n = 497] and aquaporin-4 antibody-seropositive neuromyelitis optica spectrum disorder [NMOSD, n = 308]). The Chinese normative references exhibited a later peak age than those of previously reported references, which were mainly from European and North American populations, differing by 2.3 to 9.9 years. Distinct deviations in brain structural measures were observed among individuals with neurogenerative, cerebrovascular, and neuroinflammatory diseases. We subsequently performed three clinical tasks to assess the utility of individualized deviation scores. Task 1: We estimated the individual disease propensity score relative to that of HCs, with median scores ranging from 0.84 to 0.95. Task 2: We predicted the cognitive and physical scores of individuals with neurological diseases using cross-sectional data. The correlations between the predicted and actual scores ranged from 0.13 to 0.70. Task 3: In a longitudinal cohort, we analyzed the effects of different treatments on individuals with PD. Predictions of motor outcomes in PD patients receiving medication and deep brain stimulation showed correlations between predicted and actual variables ranging from 0.16 to 0.31. In addition, we stratified individuals with MS and NMOSD according to the predicted risks of disability progression. Comparative analyses demonstrated that deviation scores outperformed raw brain structural measures in disease propensity score estimation and risk stratification of neuroinflammatory patients. Longitudinal and sensitivity analyses confirmed the stability and robustness of deviation scores in individualized brain structure quantification. Finally, using these deviation scores, we created a clinically applicable individualized brain health report. In conclusion, the quantified individualized deviation scores derived from population-specific normative references can serve as a novel approach to understanding disease pathology and contribute to accurately personalized diagnosis and prognosis for various neurological diseases.

Health sciences/Neurology/Neurological disorders

Health sciences/Health care/Diagnosis

Neurodegenerative (e.g., Alzheimer's disease [AD]), cerebrovascular (e.g., cerebral small vessel disease [CSVD]) and neuroinflammatory (e.g., multiple sclerosis [MS]) diseases are common neurological diseases with high mortality, morbidity and disability worldwide¹. Magnetic resonance imaging (MRI) has enabled the identification of brain pathology in vivo across these central nervous system (CNS) diseases, thereby guiding clinical management strategies^2,3. Quantifying MRI measures in patients with these diseases is crucial for understanding the pathogenic mechanisms, stratifying patients for targeted therapy, and assessing patient prognosis^4,5.

Previous large-scale studies using thousands of MR images have quantified and identified disease-specific structural MRI markers in various neurological diseases, elucidating the underlying pathology and providing guidance for treatment and prognosis evaluation (e.g., hippocampus atrophy in AD and thalamic atrophy in MS)^6–8. However, there are several concerns regarding previous quantified MRI measures. First, the replicability of these measures remains a major issue since they are derived mainly from standard univariate analysis frameworks with heterogeneous healthy control (HC) samples as references^4,9–11. Second, most of these studies employed classical case‒control designs and used raw MRI data to quantify disease-specific imaging markers at the group level rather than at the individual level, thereby neglecting interindividual differences and limiting the utility of these markers in guiding personalized clinical decision-making^12–14. Therefore, a personalized brain health assessment is warranted for clinical management. Finally, the influence of physiological aging on MRI measure quantification, particularly in patients with neurological disorders, poses a challenge, although this effect can be partially mitigated by employing age-matched HCs or incorporating age as an additional covariate in statistical models^4,15–17.

Recently, normative references over the lifespan have been highlighted as anchor points for standardized quantification of brain global and regional structural measures of brain disorders at the individual level^12,13,18,19. The deviation (e.g., centile score) from normative references has been regarded as a marker that is independent of the effects of brain development and aging, thus representing a novel approach to neuroimaging quantification. These normative references aid in interpreting the heterogeneity of the normal population and patients with various brain disorders, thereby facilitating more accurate disease diagnosis, the development of targeted therapy and the assessment of treatment response^12,20–23. Most previously established brain normative references were derived from European and North American populations^12,20–22. The potential differences in normative references between different populations may have been overlooked, as accumulating evidence suggests potential differences in brain morphology among different populations^24–27. Consequently, normative references used for individualized MRI feature quantification in brain disorders should be representative and population-matched²⁸.

In this study, our aim was to quantify individualized brain structural deviations, represented by centile scores, among patients with neurological diseases using Chinese population-specific normative references. We first constructed Chinese normative references and summarized deviations in patients with neurological diseases. We subsequently conducted three clinical tasks to assess the utility of these individualized deviation scores. Task 1 aimed to aid personalized diagnosis by estimating disease propensity scores (DPSs) for each patient using deviation scores with HCs as references^29,30. Task 2 aimed to assess the clinical performance of individuals by predicting baseline cognitive and physical scores of patients with neurological diseases in cross-sectional data using deviation scores. In Task 3, we evaluated treatment responses and clinical outcomes in a longitudinal cohort of patients with multiple neurological diseases, aiming to enhance treatment planning and prognostic evaluation on the basis of brain structural deviations. Finally, we developed a clinically applicable individualized brain health report.

Study Design

A total of 12,060 HCs and 3,245 patients with various neurological diseases were included. The patients included individuals with neurodegenerative (including mild cognitive impairment [MCI, n = 212], AD [n = 467], and Parkinson's disease [PD, n = 1,263]), cerebrovascular (CSVD [n = 498]), and neuroinflammatory (including MS [n = 497] and aquaporin-4 antibody-seropositive neuromyelitis optica spectrum disorder [NMOSD, n = 308]) diseases. Additional details can be found in eMethods, eFigure 1 and eFigure 2.

To quantify the individualized brain structural deviations of patients with neurological diseases, we first established a set of normative references using global and regional brain structural measures obtained from 3D T1-weighted images of healthy Chinese individuals aged 4.5 to 99 years. Briefly, a total of 228 brain structural measures were extracted at both the global and regional levels. The global measures included volumes of cortical and subcortical gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), the cerebellum, and the brainstem, along with the mean cortical thickness and total surface area. Regional measures included cortical and subcortical volume, cortical thickness, and surface area of regions determined according to the Desikan–Killiany atlas. The deviations (centile scores) of these structural measures from the established normative references were calculated for individuals in the HC and neurological disease groups.

Three clinical tasks were performed to evaluate the individualized utility of these deviation scores, as described in the Introduction. In addition to the clinical tasks, the stability of the deviation scores was assessed by comparing the baseline and follow-up MR scans. Sensitivity analyses were conducted, including (1) comparison of the fitted Chinese normative references using different models, (2) comparison of male and female normative references, and (3) comparison of the clinical tasks using other measures, including Z scores derived from the Chinese normative references and raw brain structural measures (the outputs of FreeSurfer segmentation). A concise overview of the study design is provided in Fig. 1.

Finally, we developed a clinically applicable individualized brain health report. This report incorporates the findings from the assessment of deviation scores and provides personalized insights into the structural characteristics of the brain. The report serves as a valuable tool for clinicians and patients alike, offering actionable information for improving diagnosis, treatment planning, and monitoring of neurological conditions.

Chinese Normative References and Deviations of Patients with Neurological Diseases

We first established Chinese population-specific normative references of global and regional brain structural measures using 3D T1-weighted images (eMethods, Fig. 2 and eFigure 3) obtained from 12,060 healthy Chinese individuals (age range: 4.5–99 years; mean ± SD age, 37 ± 18 years; 5,987 females [49.6%]) (Table 1). This task was accomplished with the generalized additive models for location, scale, and shape (GAMLSS) package in R. This approach parallels that of a previous study¹², in which normative references were developed on the basis of neuroimaging data from 101,457 healthy individuals primarily of European and North American descent. For clarity, we refer to these references as international references, facilitating direct comparison with the Chinese normative references established in our study. Second, we calculated deviation scores (centile scores) that benchmarked each individual scan against these normative references for both the HC and disease groups (comprising 3,245 patients with a mean age of 56 ± 16 years; 1,780 female patients [54.9%]) (Table 1). The deviation scores of the disease groups are summarized in Fig. 3. Additional analyses of these deviation scores in the disease groups using Cohen’s d, the overlapping coefficient and the univariate predictive ability are provided in eResults 1, and more details on the findings in the different neurological diseases are shown in eFigures 4–11.

Table 1

Demographic and clinical information of the included healthy controls and patients with neurological diseases.
	Overall (n = 15305)	HC (n = 12060)	MCI (n = 212)	AD (n = 467)	PD (n = 1263)	CSVD (n = 498)	MS (n = 497)	NMOSD (n = 308)	p
Age (mean (SD)), years	41.15 (19.19)	37.10 (18.03)	63.30 (9.80)	66.49 (11.60)	61.21 (10.53)	59.32 (11.82)	36.09 (12.47)	42.46 (14.21)	< 0.001
Sex (n, %)									< 0.001
Female	7767 (50.7)	5987 (49.6)	127 (59.9)	279 (59.7)	569 (45.1)	197 (39.6)	325 (65.4)	283 (91.9)
Male	7538 (49.3)	6073 (50.4)	85 (40.1)	188 (40.3)	694 (54.9)	301 (60.2)	172 (34.6)	25 (8.1)
Education years (mean (SD))	12.35 (4.58)	14.03 (4.33)	10.83 (4.26)	10.19 (4.13)	10.81 (4.49)	10.48 (3.81)	13.40 (3.39)	11.45 (3.68)	< 0.001
MMSE (mean (SD)	25.50 (5.38)	27.89 (2.73)	25.75 (3.11)	17.02 (7.62)	25.63 (4.15)	23.35 (4.77)	26.35 (6.18)	27.77 (2.48)	< 0.001
MoCA (mean (SD))	22.21 (5.74)	24.94 (3.14)	20.75 (3.94)	12.53 (7.03)	21.23 (5.03)	20.03 (5.28)	26.61 (3.16)	24.35 (4.45)	< 0.001
BVMT (mean (SD))	36.76 (14.37)	43.47 (11.94)					37.99 (14.17)	33.75 (14.44)	< 0.001
CVLT (mean (SD))	82.65 (28.94)	103.73 (22.33)					80.50 (29.18)	78.21 (27.97)	< 0.001
SDMT (mean (SD))	49.57 (16.64)	52.71 (17.36)					49.14 (13.72)	41.92 (14.90)	< 0.001
PASAT (mean (SD))	42.74 (12.89)	46.17 (13.72)					43.08 (10.50)	37.79 (12.33)	< 0.001
EDSS (median [IQR])	3.00 [1.50, 4.00]						2.00 [1.00, 3.50]	3.50 [2.00, 5.00]	< 0.001
SVD score (median [IQR])						3 [1, 4]
UPDRS-III (medication-off, mean (SD))					49.41 (17.32)
UPDRS-III (medication-on, mean (SD))					23.74 (13.49)
UPDRS-III after DBS therapy (DBS-on and medication-off, mean (SD))					27.9 (15.64)
Relapse number (median [IQR])	2.00 [2.00, 4.00]						2.00 [1.00, 3.00]	2.00 [2.00, 4.00]	0.36
Follow-up time (mean (SD)) years	3.97 (1.89)						4.27 (2.07)	3.50 (1.46)	< 0.001
Follow-up EDSS (median [IQR])	2.50 [1.50, 4.00]						2.00 [1.00, 4.00]	3.00 [1.50, 4.50]	0.01
EDSS progression, n	306						186	120
Progression (n, %)	93 (30.4)						56 (30.1)	37 (30.8)	0.96
SPMS conversion, n	186						186
Conversion (n, %)	28 (15.0)						28 (15.0)
Note: n, number; SD, standard deviation; IQR, interquartile range; HC, healthy control; AD, Alzheimer's disease; PD, Parkinson's disease; CSVD, cerebral small vessel disease; MS, multiple sclerosis; NMOSD, aquaporin-4 antibody-seropositive neuromyelitis optica spectrum disorder; MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment; BVMT, Brief Visuospatial Memory Test; CVLT, California Verbal Learning Test; PASAT, Paced Auditory Serial Addition Test; SDMT, Symbol Digit Modalities Test; UPDRS-III, Unified Parkinson's Disease Rating Scale Part III; DBS, deep brain stimulation; EDSS, Expanded Disability Status Scale; SPMS, secondary progressive multiple sclerosis. Categorical data are presented as percentages and were compared with Pearson’s chi-square test. Continuous data are presented as the means and standard deviations and were compared with one-way analysis of variance (ANOVA). Ranked data are displayed as medians and interquartile ranges and were compared with the Kruskal‒Wallis test.

For the brain global structural measures (Fig. 2), the data fit with our constructed normative references presented adjusted R-squared values ranging from 0.34 to 0.59, which were all higher than those (0.11–0.52) obtained with the calibrated international references. The Chinese normative references (aged from 4.5 to 99 years) presented increases in cortical GM volume (GMV), subcortical GMV (sGMV), WM volume (WMV), mean cortical thickness and total surface area from childhood onward, peaking at 8.2 [95% confidence interval [CI] 7.9–8.4], 17.0 [95% CI 16-17.8], 38.6 [95% CI 38.1–39], 7.0 [95% CI 6.5–7.3] and 12.5 [95% CI 10.2–13.3] years, respectively, followed by a near-linear decrease. These peaks were observed 2.3, 2.6, 9.9, 4.5 and 0.7 years later, respectively, than those of the international references. The normative curves revealed increases in the CSF volume from childhood onward and an exponential trend in the sixth decade of life, which is consistent with the international reference. Additionally, we reported the trajectories of the volumes of two structures (the cerebellum and brainstem), which were not reported in the international references. The normative curves revealed increases in the volumes of the cerebellum and brainstem from childhood onward, peaking at 14.4 [95% CI 13.4–15.4] and 30.8 [95% CI 29.5–31.9] years, respectively. Additional findings on regional structural measures are provided in eFigure 3.

We observed that patients with neurological diseases showed significant deviations from the normative references in all global measures, except for GMV, CSF, and total surface area in MCI patients; WMV in PD patients; and total surface area in CSVD patients (Fig. 3). With respect to regional deviations, we observed that MCI patients presented significant deviations in subcortical volume and cortical thickness and less involvement of cortical volume or surface area. AD patients showed significant widespread deviations in all the global and regional structural measures. PD patients showed significant deviations in subcortical volume, cortical volume and cortical thickness and less involvement of surface area. CVSD patients showed significant deviations in subcortical volume and cortical thickness and less involvement of cortical volume or surface area. MS patients presented significant widespread deviations in subcortical volume, cortical volume and cortical thickness and less involvement of surface area. NMOSD patients presented significant deviations in subcortical volume, cortical volume and surface area and less involvement of cortical thickness.

Clinical Tasks Using Individualized Deviation Scores in Patients with Neurological Diseases

Task 1: Estimation of Neurological Disease Propensity Scores by Deviation Scores

To illustrate the resemblance of deviation scores to specific diseases, we estimated the DPS (ranging from 0 to 1), reflecting the similarity between the deviation scores of HCs and those of patients with various neurological diseases. For DPS calculation, we initially employed a support vector machine (SVM) with a radial basis function kernel, following feature reduction by LASSO regression, to formulate disease-specific classification models. These models aimed to distinguish between age-, sex-, and site-matched HCs and each disease group. Model performance, measured by the area under the curve (AUC), was assessed using 10-fold cross-validation. We subsequently utilized these disease-specific models to predict individual DPSs for both HCs and patients with neurological diseases. Predictions for all the cases were conducted using 10-fold cross-validation for predictive robustness. A higher DPS generated by a disease-specific model indicates a greater likelihood of individuals being associated with that particular disease. The DPS serves as a valuable tool for disease screening in HCs and facilitating differential diagnosis of neurological diseases.

Using multivariate global and regional deviation scores, disease-specific classification models achieved AUCs of 0.82 [95% CI 0.79–0.86] for MCI, 0.92 [95% CI 0.91–0.94] for AD, 0.89 [95% CI 0.87–0.90] for PD, 0.81 [95% CI 0.78–0.84] for CSVD, 0.87 [95% CI 0.85–0.89] for MS, and 0.92 [95% CI 0.90–0.94] for NMOSD (Fig. 4a).

We utilized disease-specific classification models to estimate the DPSs for all individuals. Notably, the median estimated DPS values for each disease were as follows: 0.84 (interquartile range [IQR] 0.76–0.86) for MCI, 0.94 (IQR 0.90–0.97) for AD, 0.92 [IQR 0.87–0.94] for PD, 0.84 [IQR 0.63–0.89] for CSVD, 0.90 [IQR 0.81–0.93] for MS, and 0.95 [IQR 0.93–0.96] for NMOSD (Fig. 4a).

Task 2: Prediction of Cognitive and Physical Scores in Patients with Neurological Diseases by Deviation Scores in a Cross-sectional Design

To evaluate the ability of these deviation scores to predict disease-associated clinical variables, we employed support vector regression (SVR) with 10-fold cross-validation to assess how well the deviation scores predict baseline cognitive scores, Unified Parkinson’s Disease Rating Scale Part III (UPDRS-III) scores and Expanded Disability Status Scale (EDSS) scores across patients with various neurological diseases in cross-sectional datasets. The predictive performance was assessed by Pearson’s correlation (ranging from − 1 to 1) and the mean absolute percentage error (MAPE, ranging from 0 to +∞) between the predicted and actual variables. A higher Pearson’s correlation (r value) and lower MAPE indicate better predictive performance of the deviation scores. Furthermore, to ensure the robustness of our findings, we conducted multiple corrections using the false discovery rate (FDR) across the diseases and clinical variables to minimize the risk of false positives.

The correlations between the predicted and actual variables were between 0.25 and 0.70 for the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA) scores across HCs and all disease groups; between 0.13 and 0.67 for the Brief Visuospatial Memory Test (BVMT), California Verbal Learning Test (CVLT), Symbol Digit Modalities Test (SDMT) and Paced Auditory Serial Addition Test (PASAT) scores across HCs, MS patients and NMOSD patients; 0.24 for the UPDRS-III score in PD patients not receiving any interventions; and 0.27 and 0.42 for the EDSS score in MS patients and NMOSD patients. These correlations survived FDR correction with pFDR < 0.05, except for SDMT in patients with NMOSD, with pFDR = 0.17. The corresponding MAPEs ranged from 0.07 to 1.54. Details are provided in Fig. 4b.

Task 3: Assessment of the Treatment Response and Disease Progression of Patients with Neurological Diseases by Deviation Scores in a Longitudinal Design

To evaluate the predictive potential of deviation scores for follow-up clinical outcomes, we conducted SVM and Cox proportional hazards regression analyses for treatment response in PD patients and for stratifying MS patients and NMOSD patients with different degrees of disability progression.

(1) Predicting the Treatment Response of PD Patients Receiving medication and DBS

For PD patients, we employed an SVM with 10-fold cross-validation to predict the UPDRS-III score (n = 461) both during medication treatment and after one year of DBS therapy. The motor improvement rate represented by the UPDRS-III score, with patients who did not receive any intervention as a reference, was also assessed. The predictive model assessment procedure was the same as that in Task 2.

The correlations between the predicted and actual variables were 0.31 [95% CI 0.24–0.37] (pFDR < 0.001) and 0.22 [95% CI 0.16–0.29] (pFDR < 0.001) for the UPDRS-III score and its improvement rate when receiving medication and 0.28 [95% CI 0.19–0.36] (pFDR < 0.001) and 0.16 [95% CI 0.07–0.24] (pFDR < 0.001) for the UPDRS-III score and its improvement rate after one year of DBS therapy (Fig. 4c). The corresponding MAPEs ranged from 0.46 to 1.17.

(2) Stratifying MS and NMOSD Patients with Different Degrees of Disability Worsening at Follow-up

For MS and NMOSD patients, disability worsening, which was defined according to the difference between the baseline and follow-up (MS, n = 106, mean [SD] follow-up time = 51.24 ± 24.84 months; NMOSD, n = 82, mean [SD] follow-up time = 41.0 ± 17.52 months) EDSS scores³¹, was assessed. To stratify individuals with different degrees of disability worsening, we first conducted multivariate Cox proportional hazards regression (stepwise bidirectional elimination) using the deviation scores (conversion to secondary progressive MS from relapsing-remitting MS was also analyzed) using 10-fold cross-validation. Using the predictive relative hazard, we stratified the patients into subgroups of low- and high-risk patients. The cutoff value was the median predicted relative hazard of the corresponding disease. Kaplan‒Meier analysis with the log-rank test was conducted to investigate the potential differences in the survival curves of the stratified subgroups (Fig. 4d). These analyses for MS and NMOSD patients using deviation scores can provide information to identify patients at high risk for disease worsening and could aid in clinically stratified management of new patients. Additional analyses of the univariate and multivariate Cox proportional hazards regression results for disability worsening in MS patients and NMOSD patients are provided in eTable 1.

A Kaplan‒Meier analysis of the subgroups stratified by the predicted risk of disability worsening revealed that MS patients with a higher predicted relative hazard had a greater risk of conversion to secondary progressive MS (log-rank p = 0.0090) but had no difference in EDSS progression compared with those with a low predicted relative hazard (log-rank p = 0.99). NMOSD patients with a higher predicted relative hazard had a greater risk of disability progression than did those with a low predicted relative hazard (log-rank p = 0.0053).

Additional Stability and Sensitivity Analyses of Deviation Scores

(1) Assessment of the Stability of Deviation Scores Using Longitudinal MRI Scans

To assess the stability of the global and regional deviation scores derived from normative references and their potential role in monitoring disease development, additional longitudinal assessments of the deviation scores using independent longitudinal scans in HCs and patients with neurological diseases (follow-up times ranging from 0.5 to 3 years; HCs, n = 73; MCI, n = 42; AD, n = 56; MS, n = 42; and NMOSD, n = 32) were conducted. The IQRs of the deviation scores of the baseline and follow-up scans for each patient were used to assess the stability of the deviation scores, similar to a previous study¹². A large IQR indicated a large difference between the longitudinal scans of that patient. A linear mixed model with individual as a random effect for each deviation score in each group was conducted to monitor the disease progression reflected by the potential longitudinal changes in deviation scores.

IQRs of the longitudinal deviations across the global and local structural measures were calculated in both the HC and disease groups for patients for whom longitudinal scans were available (eResults 2 and eFigure 12). The median IQR across ages was 0.09 (IQR = [0.05, 0.16]), indicating a stable quantification of the deviation scores of these patients. Larger IQRs appeared to be more prevalent in younger and older populations but remained stable in the middle-aged population. This finding indicated that the variability in brain structural changes may be specific to different age groups. Furthermore, we analyzed the changes in the deviation scores at follow-up compared with the baseline scores. Compared with those of the baseline scans, the deviation scores of several structural measures (e.g., the left lateral orbitofrontal area in AD patients, the right lingual thickness in MS patients and the left lingual volume in NMOSD patients) were lower for the follow-up scans (p < 0.05), but the differences did not survive FDR correction, with all pFDRs > 0.05.

(2) Sensitivity Analysis of Deviation Scores in Different Models, Male and Female References, and Utilization of Other Comparative Quantitative Measures

To test whether there were differences in the normative curves using different models, disease-related findings using female and male references, and different quantitative measures, we performed sensitivity analyses to compare normative curves based on our reference and international references, calculate deviation scores using separate male and female references, and utilize other comparative quantitative measures (Z scores derived from the normative references and raw brain structural measures).

For the normative reference comparisons (eResults 3 and eFigure 13), we assessed the consistency of our normative references (median value curves) with international references, as well as the site-calibrated international references, using Pearson’s correlation and mean absolute scaled error (MASE). Our normative references, international references and site-calibrated references showed high consistency, with a minimum correlation greater than 0.9 and the lowest correlation for the WMV curves. The normative references of cortical thickness showed the largest MASE, indicating a large difference between our normative reference and the international reference.

The normative references stratified by sex are presented in eResults 4 and eFigures 14–16. Compared with females, healthy male populations presented greater median values of brain volume, cortical thickness and total surface area across the lifespan, and supratentorial brain tissue volume and total surface area were significant. Comparisons between disease patients and HCs stratified by sex revealed similar global and regional deviation scores across diseases. However, the global and regional deviations of the male subgroups of MCI, MS and NMOSD patients were less statistically significant.

In the comparison of different measures (eResults 5 and eFigures 17–24), the analyses revealed that these indices had comparable abilities for the detection of measures that were statistically significant between disease groups and HCs. More than 60% of the detected MRI measures overlapped with these quantitative measures, implying that a combination of these methods may provide robust brain structural markers in patients with neurological diseases. For DPS estimation, the disease-specific classification model focused more on the corresponding disease group using the centile score than on the raw brain structural measures (eFigure 22). The centile score outperformed the raw brain structural measures in stratifying MS and NMOSD patients with different degrees of disability worsening (eFigure 24).

Clinically Applicable Brain Health Report for Individuals by Deviation Scores

Finally, we developed a clinically applicable brain health report for individuals on the basis of the global and regional deviation scores (Fig. 5). In this brain health report, we first provided global deviation scores for GMV, sGMV, WMV and cerebellum and brainstem volumes in the framework of normative references. We then provided regional deviation score maps of subcortical and cortical volumes. The summarized tables for these global and regional deviation scores are provided, where deviation scores lower than 5% or higher than 95% of the normative references are highlighted. Additionally, we provided the DPS as a reference for clinical diagnosis.

In this study, we established population-specific normative references of brain structural measures using a large cohort of healthy Chinese individuals, quantified the deviation scores of patients with neurological diseases and investigated the clinical utility of these scores. First, we observed that the Chinese normative references exhibited a later peak age compared with those of the international references predominantly derived from European and North American populations, differing by 2.3 to 9.9 years. We also observed that patients with neurological diseases exhibited disease-specific deviations in global and regional brain structural measures. The deviation scores displayed promising utility in estimating disease propensity with high accuracy and demonstrated the ability to predict cognitive and physical performance in disease groups. Furthermore, these scores exhibited prognostic value in predicting the response to medication and DBS treatment in PD patients and disability progression in MS and NMOSD patients. Notably, the deviation scores outperformed the raw structural measures in clinical tasks, including DPS estimation and stratifying MS and NMOSD patients with different degrees of disability progression, while they were comparable to the Z scores. Stability analyses using follow-up scans and sensitivity analyses confirmed the robustness of the deviation scores in representing brain structural measures. Finally, we developed a clinically applicable report for individuals, leveraging quantitative deviation scores to provide insights into brain health.

In our study, while the majority of measures in the normative references from a healthy Chinese population were consistent with those of the international references, distinct trajectories were observed for certain structural measures, such as WMV and cortical thickness^12,24,25. These disparities, particularly in cortical thickness, are corroborated by previously reported significant morphological differences between Asian and Caucasian populations^25–27,32. Even after calibrating for site effects in the international references, these differences persisted, suggesting that these calibrated references may not be optimal for populations that significantly diverge from those included in the original datasets. Differences in sample sizes may result in discrepancies between our established normative references and the international references. Despite the use of a smaller sample size, our normative references demonstrated better alignment with Chinese data than the international references did. These findings highlight the critical need for constructing population-specific normative references in neuroscience and clinical neurology research; offering a series of tailored normative references for brain structural measures that are specific to the Chinese population could increase the accuracy and relevance of studies involving Chinese individuals.

The normative reference-derived centile scores are considered the optimal index because they are independent of the assumption of a normal data distribution, as supported by previous research^12,18. We summarized the deviations in global and regional structural measures in patients with different neurological diseases, and different deviation patterns were observed. MCI patients presented lower deviation scores for a few subcortical volumes (e.g., hippocampus volume) and cortical thicknesses (e.g., cingulate), indicating that these measures are sensitive to brain atrophy in individuals with MCI^33,34. AD patients showed lower deviations across all the global and regional measures, suggesting severe brain atrophy in AD patients^1511,35. PD patients also had lower subcortical volume, cortical volume and cortical thickness, but the deviation was relatively moderate compared with that of AD patients^7,11,36,37. Compared with patients with neurodegenerative diseases, patients with CSVD had lower deviations in subcortical volume (predominantly in the thalamus) and cortical thickness, implying disease-specific atrophy^38,39. Both MS and NMOSD patients exhibited widespread lower deviations of global and regional structural measures; however, more areas were involved in MS patients than in NMOSD patients, indicating that brain atrophy in MS patients is more severe than that in NMOSD patients^8,40–42. These findings demonstrated distinct patterns of brain structural deviations in neurodegenerative, cerebrovascular and neuroinflammatory diseases related to their specific disease pathology, which is consistent with previous findings from classical case‒control comparisons^4,42. By utilizing the same normative references, deviation scores provide a comprehensive view of brain structural changes across neurological diseases, facilitating comparisons between diseases.

In this study, we evaluated the individualized utility of deviation scores in common clinical tasks. We first illustrated the ability of calculated deviation scores to estimate the disease propensity of neurological conditions. The disease-specific classification model for disease propensity prediction exhibited competitive performance for the studied diseases, suggesting its potential as a clinically applicable tool for aiding in the differential diagnosis of neurological diseases and screening the healthy population for high neurological disease risk. In our study, deviation scores outperformed raw structural measures, as raw measures also demonstrated relatively high DPSs for non-targeted diseases and HCs. These results suggest that deviation scores may be more closely linked to disease pathology, yielding higher predictive values for patients with similar pathologies and lower values for patients with distinct pathologies. Moreover, raw measures encompass not only disease pathology but also other effects, such as age and sex, despite our use of age- and sex-matched HCs as references. This finding underscores the potential value of deviation scores in elucidating a more "confident" underlying pathology than that predicted by raw measures, but additional validations are necessary.

The quantified deviation scores reflecting brain pathology have demonstrated close associations with cognitive performance across various diseases and normal aging^{16,43 44}. Additionally, these deviation scores are associated with physical disability in neuroinflammatory diseases and serve as objective markers for motor dysfunction in PD patients^36,45. Moreover, the quantified deviation scores show promise in predicting motor improvement in PD patients receiving medication and DBS therapy. The scores also exhibit high predictive performance for disability progression in MS and NMOSD patients. Furthermore, deviation scores outperformed raw measures in stratifying MS and NMOSD patients with different disease progression risks. These findings suggest that quantified deviation scores could serve as valuable tools for monitoring cognitive and physical deterioration in both the aging population and patients with neurological diseases. Additionally, they have the potential to enhance predictions of treatment response and disease progression across various neurological conditions.

Finally, the proposed clinically applicable brain health report utilizing deviation scores has the potential to offer objective and automated assessments of brain health. This report could aid in screening high-risk populations and distinguishing between different neurological diseases. This innovation has the potential for widespread adoption in clinical practice.

There are several limitations in this study. First, most data were retrospectively collected from multiple sites, and data heterogeneity is a major limitation for normative reference construction. Even if we used site as a random effect to remove the site effects in GAMLSS, potential bias may still be present. Second, in the present study, we constructed a lifespan trajectory with a minimum age of 4.5 years and had no available data for infants aged less than two years, who experience rapid and dynamic development in terms of both neuron complexity and white matter myelination⁴⁶. This lack of data may bias the fitting trajectories, given the differences with those in the international references, especially for cortical thickness. Future works involving more younger individuals could increase confidence in the current Chinese population-specific normative references. Third, as an intrinsic limitation of retrospective studies, we collected data on the maximum number of available diseases without statistical estimation. Additionally, most of the patients were recruited from local institutes, and external validation using patients from other sites may support the application of this model and increase the confidence of the clinical findings. Fourth, longitudinal MR scans were available for only a subset of cases numbering less than 300, with limited follow-ups of less than three years and fewer than five time points. Therefore, the longitudinal stability assessment may be biased, as there was not enough follow-up time points, and additional validation is needed. Finally, although we assessed the treatment response for PD patients and disability progression for MS and NMOSD patients, information for MCI, AD and CSVD patients was limited. Studies with larger samples with standardized treatments for these diseases could confirm the utility of the current findings for clinical application even in other diseases.

In this study, we proposed individualized quantitative deviation scores as novel imaging markers for various neurological diseases using Chinese population-specific normative references derived from 3D T1-weighted MR images. Deviation scores can be used to reveal disease pathology, estimate disease propensity, predict cognitive and physical performance, and evaluate disease prognosis across neurological diseases; these scores exhibited superior performance compared with raw measures predominantly in diagnostic and prognostic tasks.

Participants and Study Design

This study was approved by the ethics committees of Beijing Tiantan Hospital, Capital Medical University, Beijing, China (Nos. KY-2019-050-02 and KY-2019-140-02). The participants or their guardians provided written informed consent. Ages or birth dates were extracted from the metadata of DICOM files or were reported by participants or their guardians. At the time of data analysis, 12,060 HCs and 3,245 patients with neurological diseases, including 212 with MCI, 467 with AD, 1263 with PD, 498 with CSVD, 497 with MS (in this study, we used patients with relapsing–remitting MS) and 308 with NMOSD with aquaporin-4 antibody seropositivity, participated in this study. Clinical and MRI data were collected from 44 sites in China (see details in eFigure 1 and eFigure 2). The MR images of the included participants passed an automatic quality check using Euler's number and were visually checked by Z.Z. (an MR physicist with more than 10 years of experience in MR neuroimaging). The inclusion and exclusion criteria are available in eMethods 1.

This multicenter study aimed to quantify deviation scores in patients with neurological diseases using normative references and assess the clinical utility of these scores in various clinical tasks. A brief description of the study design is provided in Fig. 1. First, we established normal references of global and regional structural measures derived from 3D T1-weighted images obtained from 12,060 HCs and calculated the deviation scores (centile scores) of participants with neurological diseases. We additionally compared the established normal references with previously reported international references and site-calibrated references¹². Deviations in neurodegenerative, cerebrovascular and neuroinflammatory diseases were summarized.

We performed three downstream clinical tasks using the deviation scores. Task 1: We estimated the DPS, which is defined as the predicted disease probability, using the global and regional deviation scores of patients with HCs as references. Task 2: To assess the predictive values of the deviation scores, we assessed the ability of the deviation scores to predict the cognitive and physical scores in patients with neurological diseases and HCs. Task 3: To explore whether these deviation scores have clinical prognostic value, we assessed their ability to predict prognosis in PD patients based on the treatment response to medication and DBS therapy and in MS and NMOSD patients based on disability worsening. Finally, additional longitudinal MR scans were used to assess the stability of deviation scores between the baseline and follow-up MR scans, and sensitivity analyses on normative curves using different models, sex stratification, and other quantitative measures (Z scores derived from the normative references and raw brain structural measures) were conducted. We also developed a clinically applicable brain health report for individuals using the deviation scores.

This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Data Collection and Processing

The available clinical variables, including age, sex and education years (n = 3,710), were collected (Table 1). The cognitive tests included the MMSE (n = 2,551) and MoCA (n = 2,401). Other cognitive tests with available results for MS and NMOSD patients included the BVMT (n = 249), CVLT (n = 259), SDMT (n = 488) and PASAT (n = 466). Subsets of PD patients without medication (n = 771) and with medication (n = 766) and DBS therapy (n = 461) completed UPDRS-III assessments. A subset of patients with neuroinflammatory diseases completed physical disability assessments using the EDSS. A subset of PD patients (n = 461) had one-year follow-up scores on the UPDRS-III after DBS therapy. A subset of patients with neuroinflammatory diseases completed a follow-up EDSS (MS, n = 106, mean [SD] follow-up time = 51.24 ± 24.84 months; NMOSD, n = 82, mean [SD] follow-up time = 41.0 ± 17.52 months), and EDSS progression in these patients was assessed³¹. A subset of MS patients (n = 106) with clinical follow-up data on the conversion of relapsing-remitting MS to secondary progressive MS were additionally assessed³¹. Available follow-up data (follow-up times ranging from 0.5 to 3 years) from MR scans of HCs (n = 73), MCI patients (n = 42), AD patients (n = 56), MS patients (n = 42) and NMOSD patients (n = 32) were collected for longitudinal assessments of the normative reference-derived deviation scores. All MR images were reviewed by two experienced raters (Z.Z. with 8 years of experience in MR physics and Y.D. with 15 years of experience in neuroradiology) for image quality control. Disagreements were resolved by a third rater (Y.L. with 20 years of experience in neuroradiology).

Automated structural MRI segmentation of 3D T1-weighted images was conducted using the cortical and subcortical parcellation stream of FreeSurfer Software Suite, version 7.0 (Laboratory for Computational Neuroimaging at the Athinoula A. Martinos Center for Biomedical Imaging), which is based on the Desikan‒Killiany atlas (eMethods 2). Segmentations were visually inspected and edited (if necessary) by Z.Z. and then assessed by the Euler number (an index to represent the image segmentation quality using FreeSurfer; outliers were defined as the 1.5 interquartile range (IQR) below the first quartile (Q1) [Q1-1.5*IQR])⁴⁷.

We quantified deviation scores (centile scores) for each structural measure from those of the normative references (n = 12,060) using the GAMLSS package in R, similar to the international references in a previous study¹². We used HC data to fit the normative references for the Chinese population and to calibrate the international references for site effects as described previously¹². Age and sex were regarded as fixed effects, and site-specific protocols were regarded as random effects in the GAMLSS package. FreeSurfer was also used for the international reference calibrations. Additionally, 10-fold cross-validation was used to calculate the deviation scores for HCs to ensure that the predicted deviation scores of each individual were not included in the normative model training.

Statistical analysis

Statistical analysis was conducted with R software (version 4.1.3, R Foundation for Statistical Computing, Vienna, Austria). Categorical data are presented as percentages and were compared with Pearson’s chi-square test. Continuous data are displayed as the means and SDs and were compared with one-way analysis of variance (ANOVA). Ranked data are displayed as medians and IQRs and were compared with the Kruskal‒Wallis test.

For each deviation score, a Monte Carlo test between patients and age-, sex- and site-matched (if available, otherwise unmatched) HCs was performed. This approach mirrors the traditional mass-univariate approach used in neuroimaging by estimating an independent model for each variable. In addition, as shown in the Supplementary materials, the variable showing the strongest effect (largest Cohen’s d) was selected, which was calculated for the group factor (HC versus disease). Correction for multiple comparisons using the FDR was performed within each disease subset across the global and regional deviation scores. Furthermore, we calculated the overlapping coefficient of the brain structural deviation distribution for the maximum effect variables across individuals in the disease and HC groups to illustrate the dissimilarity of the structural deviations between patients with neurological diseases and HCs. To quantify their predictive potential, univariate logistic regression with 10-fold cross validation was performed to classify patients and HCs, and the accuracy and AUCs were calculated. Additionally, Cohen’s d and the distributional overlapping coefficient between each pair of disease groups were calculated.

An SVM with 10-fold cross-validation after feature reduction by least absolute shrinkage and selection operator (LASSO) regression was used to develop disease-specific classification models (classification between each disease group and HCs) and then used to predict individual DPSs. The classification model performance (AUC) was evaluated using 10-fold cross-validation. Multivariate SVR with 10-fold cross-validation was used to assess the predictive ability of deviation scores for cognitive scores (in this analysis, education was not considered, as a majority of the patients had no available education information), UPDRS-III scores in PD patients and EDSS scores in MS and NMOSD patients. The predictive performance was assessed by the Pearson correlation coefficient (ranging from − 1 to 1) and MAPE (ranging from 0 to +∞) between the predicted and actual variables. Correction for multiple comparisons using the FDR was performed across diseases and clinical variables.

To assess whether the deviation scores had prognostic value for follow-up clinical outcomes, we performed SVR and Cox proportional hazards regression for treatment response in PD patients and disease progression in MS and NMOSD patients, respectively. Using SVR with 10-fold cross-validation, we assessed the UPDRS III scores of PD patients receiving medication and after one year of DBS therapy, with improvement in the UPDRS III score in patients without any intervention as the reference, using deviation scores. The predictive performance assessment was the same as that of the above predictive tasks for the cognitive and physical scores. For MS and NMOSD patients, disability progression, which was determined based on the baseline and follow-up EDSS scores³¹, was assessed by multivariate Cox proportional hazards regression (stepwise bidirectional elimination) using deviation scores. The predicted relative hazard for each case was determined by 10-fold cross-validation to ensure that the predicted case was not included in the model training. Using the predictive relative hazard, we stratified the patients into subgroups of low- and high-risk patients. The cutoff value was the median predicted relative hazard of the corresponding disease. Kaplan‒Meier analysis with the log-rank test was conducted to investigate the potential differences in the survival curves of the stratified subgroups. For MS patients, conversion to secondary progressive MS conversion from relapsing-remitting MS³¹ was also assessed using the above methods.

All P values were two-sided, and significance was set at P < 0.05. The data were analyzed from December 29, 2022, to November 29, 2023, using R, version 4.2.

Additional stability assessment and sensitivity analysis

Additional stability assessments of the deviation scores for the baseline and follow-up longitudinal scans in HCs and patients with neurological diseases were conducted using IQRs and a linear mixed model with individual as a random effect for each deviation score in each group. We further performed a series of sensitivity analyses, including (1) comparisons with international references, (2) normative references and deviations of disease groups stratified by sex, and (3) comparisons of Z scores and raw brain structural measures (the outputs of FreeSurfer segmentation).

All the codes used for the statistical analyses and figures are publicly available at GitHub: https://github.com/zhuozhizheng/Chinese_normative_reference_and_downstream_clinical_applications.

Acknowledgments

We acknowledge all the contributors who provided MR data from healthy volunteers.

Funding

This work was supported by the National Science Foundation of China (NO. 822020840) (ZZ), Beijing Municipal Natural Science Foundation for Distinguished Young Scholars (No. JQ20035) (YL), Ministry of Science and Technology of China-National Key Research Project (2022YFC2009904) (YL), Beijing Hospital Management Center-Climb Plan (YL), Beijing Young Scholars (YL), Capital Medical University Young Scholars (YL), and Young Scientists Program of Beijing Tiantan Hospital, Capital Medical University (YSP202205) (ZZ).

Competing interests

Jinyuan Weng is an employee of Philips Healthcare, Shanghai, China.

The other authors declare that they have no competing interests.

Data availability

Source data and main codes for the statistical analysis and figures are available at GitHub: https://github.com/zhuozhizheng/Chinese_normative_reference_and_downstream_clinical_applications.

Author contributions

Study design: Zhizheng Zhuo, Xi-Nian Zuo and Yaou Liu.

Data analysis: Zhizheng Zhuo, Yinshan Wang and Jinyuan Weng.

Statistical analysis: Zhizheng Zhuo, Xiaolu Xu and Li Chai.

Data collection: Zhizheng Zhuo, Xiaolu Xu, Li Chai, Yinshan Wang, Yuna Li, Jun Xu, Yulu Shi, Decai Tian, Yuteng Bai, Jinaguo Zhang, Jianrui Li, Zhiqiang Zhang, Fuqing Zhou, Hui Dai and Yunyun Duan.

Manuscript writing: Zhizheng Zhuo and Xiaolu Xu.

Manuscript editing: Li Chai, Min Guo, Dan Cheng, Siyao Xu, Yuna Li, James Cole, ‪Xi-Nian Zuo and Yaou Liu.

Collaborators, G. B. D. N. S. D. Global, regional, and national burden of disorders affecting the nervous system, 1990–2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet Neurol 23, 344–381, doi:10.1016/S1474-4422(24)00038-3 (2024).
Deuschl, G. et al. The burden of neurological diseases in Europe: an analysis for the Global Burden of Disease Study 2017. Lancet Public Health 5, e551-e567, doi:10.1016/S2468-2667(20)30190-0 (2020).
Group, G. B. D. N. D. C. Global, regional, and national burden of neurological disorders during 1990–2015: a systematic analysis for the Global Burden of Disease Study 2015. Lancet Neurol 16, 877–897, doi:10.1016/S1474-4422(17)30299-5 (2017).
Risacher, S. L. & Saykin, A. J. Neuroimaging in aging and neurologic diseases. Handb Clin Neurol 167, 191–227, doi:10.1016/B978-0-12-804766-8.00012-1 (2019).
Cauda, F. et al. Brain structural alterations are distributed following functional, anatomic and genetic connectivity. Brain 141, 3211–3232, doi:10.1093/brain/awy252 (2018).
van Oostveen, W. M. & de Lange, E. C. M. Imaging Techniques in Alzheimer's Disease: A Review of Applications in Early Diagnosis and Longitudinal Monitoring. Int J Mol Sci 22, doi:10.3390/ijms22042110 (2021).
Prange, S., Metereau, E. & Thobois, S. Structural Imaging in Parkinson's Disease: New Developments. Curr Neurol Neurosci Rep 19, 50, doi:10.1007/s11910-019-0964-5 (2019).
Sastre-Garriga, J., Pareto, D. & Rovira, A. Brain Atrophy in Multiple Sclerosis: Clinical Relevance and Technical Aspects. Neuroimaging Clin N Am 27, 289–300, doi:10.1016/j.nic.2017.01.002 (2017).
Lewis, A. S. & Nobis, W. P. Narrowing the translational research gap by aligning replication concepts in basic and clinical neuroscience. J Neurosci Methods 392, 109878, doi:10.1016/j.jneumeth.2023.109878 (2023).
Botvinik-Nezer, R. & Wager, T. D. Reproducibility in Neuroimaging Analysis: Challenges and Solutions. Biol Psychiatry Cogn Neurosci Neuroimaging 8, 780–788, doi:10.1016/j.bpsc.2022.12.006 (2023).
Young, P. N. E. et al. Imaging biomarkers in neurodegeneration: current and future practices. Alzheimers Res Ther 12, 49, doi:10.1186/s13195-020-00612-7 (2020).
Bethlehem, R. A. I. et al. Brain charts for the human lifespan. Nature 604, 525–533, doi:10.1038/s41586-022-04554-y (2022).
Rutherford, S. et al. Evidence for embracing normative modeling. Elife 12, doi:10.7554/eLife.85082 (2023).
Marquand, A. F., Rezek, I., Buitelaar, J. & Beckmann, C. F. Understanding Heterogeneity in Clinical Cohorts Using Normative Models: Beyond Case-Control Studies. Biol Psychiatry 80, 552–561, doi:10.1016/j.biopsych.2015.12.023 (2016).
Pini, L. et al. Brain atrophy in Alzheimer's Disease and aging. Ageing Res Rev 30, 25–48, doi:10.1016/j.arr.2016.01.002 (2016).
Fletcher, E. et al. Brain volume change and cognitive trajectories in aging. Neuropsychology 32, 436–449, doi:10.1037/neu0000447 (2018).
MacDonald, M. E. & Pike, G. B. MRI of healthy brain aging: A review. NMR Biomed 34, e4564, doi:10.1002/nbm.4564 (2021).
Rutherford, S. et al. The normative modeling framework for computational psychiatry. Nat Protoc 17, 1711–1734, doi:10.1038/s41596-022-00696-5 (2022).
Fraza, C. J., Dinga, R., Beckmann, C. F. & Marquand, A. F. Warped Bayesian linear regression for normative modelling of big data. Neuroimage 245, 118715, doi:10.1016/j.neuroimage.2021.118715 (2021).
Verdi, S. et al. Personalising Alzheimer's Disease progression using brain atrophy markers. medRxiv, doi:10.1101/2023.06.15.23291418 (2023).
Verdi, S. et al. Revealing Individual Neuroanatomical Heterogeneity in Alzheimer Disease Using Neuroanatomical Normative Modeling. Neurology 100, e2442-e2453, doi:10.1212/WNL.0000000000207298 (2023).
Worker, A. et al. Extreme deviations from the normative model reveal cortical heterogeneity and associations with negative symptom severity in first-episode psychosis from the OPTiMiSE and GAP studies. Transl Psychiatry 13, 373, doi:10.1038/s41398-023-02661-6 (2023).
Italinna, V., Kaltiainen, H., Forss, N., Liljestrom, M. & Parkkonen, L. Using normative modeling and machine learning for detecting mild traumatic brain injury from magnetoencephalography data. PLoS Comput Biol 19, e1011613, doi:10.1371/journal.pcbi.1011613 (2023).
Turney, I. C. et al. Brain Aging Among Racially and Ethnically Diverse Middle-Aged and Older Adults. JAMA Neurol 80, 73–81, doi:10.1001/jamaneurol.2022.3919 (2023).
Tang, Y. et al. Brain structure differences between Chinese and Caucasian cohorts: A comprehensive morphometry study. Hum Brain Mapp 39, 2147–2155, doi:10.1002/hbm.23994 (2018).
Bhalerao, G. V. et al. Construction of population-specific Indian MRI brain template: Morphometric comparison with Chinese and Caucasian templates. Asian J Psychiatr 35, 93–100, doi:10.1016/j.ajp.2018.05.014 (2018).
Kang, D. W. et al. Differences in cortical structure between cognitively normal East Asian and Caucasian older adults: a surface-based morphometry study. Sci Rep 10, 20905, doi:10.1038/s41598-020-77848-8 (2020).
Rosenbaum, P. R. Impact of multiple matched controls on design sensitivity in observational studies. Biometrics 69, 118–127, doi:10.1111/j.1541-0420.2012.01821.x (2013).
Williamson, E. J. & Forbes, A. Introduction to propensity scores. Respirology 19, 625–635, doi:10.1111/resp.12312 (2014).
Groenwold, R. H. [Propensity scores in observational research]. Ned Tijdschr Geneeskd 157, A6179 (2013).
Zhuo, Z. et al. Subtyping relapsing-remitting multiple sclerosis using structural MRI. J Neurol 268, 1808–1817, doi:10.1007/s00415-020-10376-7 (2021).
Wu, J. et al. Age-specific structural fetal brain atlases construction and cortical development quantification for chinese population. Neuroimage 241, 118412, doi:10.1016/j.neuroimage.2021.118412 (2021).
Raine, P. J. & Rao, H. Volume, density, and thickness brain abnormalities in mild cognitive impairment: an ALE meta-analysis controlling for age and education. Brain Imaging Behav 16, 2335–2352, doi:10.1007/s11682-022-00659-0 (2022).
Sambuchi, N., Geda, Y. E. & Michel, B. F. Cingulate cortex in pre-MCI cognition. Handb Clin Neurol 166, 281–295, doi:10.1016/B978-0-444-64196-0.00015-7 (2019).
Wu, B. S. et al. Cortical structure and the risk for Alzheimer's disease: a bidirectional Mendelian randomization study. Transl Psychiatry 11, 476, doi:10.1038/s41398-021-01599-x (2021).
Filippi, M. et al. Progressive brain atrophy and clinical evolution in Parkinson's disease. Neuroimage Clin 28, 102374, doi:10.1016/j.nicl.2020.102374 (2020).
Burton, E. J., McKeith, I. G., Burn, D. J., Williams, E. D. & O'Brien, J. T. Cerebral atrophy in Parkinson's disease with and without dementia: a comparison with Alzheimer's disease, dementia with Lewy bodies and controls. Brain 127, 791–800, doi:10.1093/brain/awh088 (2004).
Li, H. et al. Regional cortical thinning, demyelination and iron loss in cerebral small vessel disease. Brain 146, 4659–4673, doi:10.1093/brain/awad220 (2023).
Li, H. et al. Dissociable Contributions of Thalamic-Subregions to Cognitive Impairment in Small Vessel Disease. Stroke 54, 1367–1376, doi:10.1161/STROKEAHA.122.041687 (2023).
Kim, S. H. et al. Widespread cortical thinning in patients with neuromyelitis optica spectrum disorder. Eur J Neurol 23, 1165–1173, doi:10.1111/ene.13011 (2016).
Minagar, A. et al. The thalamus and multiple sclerosis: modern views on pathologic, imaging, and clinical aspects. Neurology 80, 210–219, doi:10.1212/WNL.0b013e31827b910b (2013).
Manogaran, P. et al. Optical Coherence Tomography and Magnetic Resonance Imaging in Multiple Sclerosis and Neuromyelitis Optica Spectrum Disorder. Int J Mol Sci 17, doi:10.3390/ijms17111894 (2016).
Armstrong, N. M. et al. Associations between cognitive and brain volume changes in cognitively normal older adults. Neuroimage 223, 117289, doi:10.1016/j.neuroimage.2020.117289 (2020).
Arvanitakis, Z., Shah, R. C. & Bennett, D. A. Diagnosis and Management of Dementia: Review. JAMA 322, 1589–1599, doi:10.1001/jama.2019.4782 (2019).
Heinen, R. et al. Small vessel disease lesion type and brain atrophy: The role of co-occurring amyloid. Alzheimers Dement (Amst) 12, e12060, doi:10.1002/dad2.12060 (2020).
Gilmore, J. H., Knickmeyer, R. C. & Gao, W. Imaging structural and functional brain development in early childhood. Nat Rev Neurosci 19, 123–137, doi:10.1038/nrn.2018.1 (2018).
Monereo-Sanchez, J. et al. Quality control strategies for brain MRI segmentation and parcellation: Practical approaches and recommendations - insights from the Maastricht study. Neuroimage 237, 118174, doi:10.1016/j.neuroimage.2021.118174 (2021).

Yes there is potential Competing Interest. Jinyuan Weng is an employee of Philips Healthcare, Shanghai, China. The other authors declare that they have no competing interests.

Supplementmaterialsnaturemedicine.docx
Supplementary Materials

Download PDF

Version 1

posted

You are reading this latest preprint version

Quantifying individualized deviations of brain structure in patients with multiple neurological diseases from normative references

Status:

Version 1

Abstract

Figures

Introduction

Results

Study Design

Chinese Normative References and Deviations of Patients with Neurological Diseases

Clinical Tasks Using Individualized Deviation Scores in Patients with Neurological Diseases

Task 1: Estimation of Neurological Disease Propensity Scores by Deviation Scores

(1) Predicting the Treatment Response of PD Patients Receiving medication and DBS

(2) Stratifying MS and NMOSD Patients with Different Degrees of Disability Worsening at Follow-up

Additional Stability and Sensitivity Analyses of Deviation Scores

(1) Assessment of the Stability of Deviation Scores Using Longitudinal MRI Scans

Clinically Applicable Brain Health Report for Individuals by Deviation Scores

Discussion

Conclusion

Methods

Participants and Study Design

Data Collection and Processing

Statistical analysis

Additional stability assessment and sensitivity analysis

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1