Multimodal diffusion MRI detects microstructural abnormalities within emotion- and cognition-related circuits, particularly the amygdala and cingulate cortex, with NODDI parameters in the left putamen showing the highest AUC (0.916) for distinguishing PSD from healthy controls and DKI metrics in the left anterior cingulate showing the highest AUC (0.916) for differentiating non-PSD stroke patients from healthy controls.
Key Findings
Results
The left anterior cingulate DKI MD_p10 demonstrated the best performance for distinguishing healthy controls from non-PSD stroke patients.
AUC = 0.916 for HC vs. non-PSD discrimination
DKI (Diffusion Kurtosis Imaging) was the model yielding the top metric for this group comparison
The metric used was the 10th percentile of mean diffusivity (MD_p10) within the left anterior cingulate gyrus
This finding suggests anterior cingulate microstructural changes occur even in stroke patients without depression
Results
The left putamen NODDI ODI_p10 showed the highest AUC for distinguishing healthy controls from PSD patients.
AUC = 0.916 for HC vs. PSD discrimination
Metric was the 10th percentile of Orientation Dispersion Index (ODI) from NODDI model in the left putamen
NODDI (Neurite Orientation Dispersion and Density Imaging) parameters in temporolimbic regions showed relatively higher AUC values in distinguishing PSD from HC
The putamen is part of the striatum and is involved in emotion-related circuits
Results
The left putamen NODDI ODI_p50 yielded the highest AUC for distinguishing non-PSD stroke patients from PSD patients.
AUC = 0.856 for non-PSD vs. PSD discrimination
Metric was the 50th percentile (median) of ODI from NODDI model in the left putamen
This represents the most clinically relevant comparison for early PSD detection among stroke patients
The lower AUC compared to the HC vs. PSD comparison (0.916) suggests greater difficulty discriminating PSD from non-PSD stroke than from healthy individuals
Results
Altered gray matter microstructure in PSD and non-PSD stroke patients was primarily identified in the middle frontal gyrus, posterior cingulate gyrus, paracingulate gyrus, and amygdala.
These regions are components of emotion- and cognition-related circuits
Four diffusion models were compared: DTI, DKI, MAP, and NODDI
Diffusion spectrum imaging was used for data acquisition
Infarct location was not included as a covariate, which the authors acknowledge as a limitation
Results
After FDR correction, multiple diffusion metrics were significantly correlated with MMSE and MoCA scores, but no imaging parameter showed a significant correlation with HAMD-21 depression scores.
Partial correlations were adjusted for age, sex, and years of education
MMSE (Mini-Mental State Examination) and MoCA (Montreal Cognitive Assessment) are cognitive measures
HAMD-21 is the Hamilton Depression Rating Scale used to assess depression severity
The absence of significant correlation between diffusion metrics and HAMD-21 after FDR correction suggests the imaging markers may better capture cognitive rather than purely depressive microstructural changes
Methods
Significant group differences in age, years of education, and all clinical scale scores were observed among the three participant groups.
Three groups: 23 healthy controls (HC), 26 non-PSD stroke patients (HAMD-21 < 7), and 15 PSD patients (HAMD-21 ≥ 7)
All intergroup differences in clinical scales reached p < 0.05
Groups were age- and sex-matched
Clinical assessments included HAMD-21, ADL (Activities of Daily Living), MoCA, and MMSE
The relatively small sample size, especially in the PSD group (n=15), is a notable methodological limitation
Results
Among the four diffusion models compared, NODDI parameters showed relatively higher diagnostic performance for PSD detection while DKI metrics showed relatively higher performance for detecting stroke-related changes in non-depressed patients.
Four models compared were DTI (Diffusion Tensor Imaging), DKI (Diffusion Kurtosis Imaging), MAP (Mean Apparent Propagator), and NODDI
Temporolimbic NODDI parameters showed relatively higher AUC values in distinguishing PSD from HC
Anterior cingulate DKI metrics showed relatively higher AUC values for differentiating non-PSD stroke patients from HC
The authors suggest combining multiple diffusion models may aid in the identification of PSD
What This Means
This research suggests that advanced brain imaging techniques can detect subtle changes in the brain's gray matter that may help identify post-stroke depression (PSD) earlier and more objectively. The study compared four different mathematical models for analyzing brain diffusion MRI scans in three groups: healthy people, stroke patients without depression, and stroke patients with depression. Key brain regions showing abnormalities included the middle frontal gyrus, posterior cingulate gyrus, paracingulate gyrus, and amygdala — areas known to be involved in emotion and thinking. Notably, a measure called NODDI ODI in the left putamen achieved an AUC (a measure of diagnostic accuracy, where 1.0 is perfect) of 0.916 for distinguishing depressed stroke patients from healthy controls, and 0.856 for distinguishing depressed from non-depressed stroke patients.
Interestingly, different imaging models appeared to capture different aspects of stroke-related brain changes. The DKI model performed best at detecting changes in non-depressed stroke patients (in the anterior cingulate region), while the NODDI model performed best at identifying changes specific to post-stroke depression (in the putamen). The brain changes detected by these imaging metrics correlated with cognitive test scores but not with depression severity scores, suggesting these MRI measures may be more sensitive to thinking-related brain changes than to mood-related changes specifically.
This research suggests that using multiple diffusion MRI models together, rather than relying on a single standard approach, could improve the early detection of post-stroke depression — a condition that is frequently missed in clinical practice. However, the study had a small sample size (only 15 PSD patients) and did not account for the location of each patient's stroke, which are important limitations that mean these findings need to be confirmed in larger, more controlled studies before they could be applied clinically.
Zhang F, Zhang J, Zhang L, Jin H, Li D, Zhao W. (2026). Comparative evaluation of diffusion models for detecting gray matter microstructural alterations in post-stroke depression patients.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1831557