MRI structural and functional axial asymmetry in the brain-first versus body-first subtypes of Parkinson’s disease Authors: Jiaxin Shi, Heng Zhang, Xufeng Wang, Huimin Sun, Xingyue Cao, Caiting Gan, Qianqian Si, Shiyi Ye, Zhuang Zhu, Chenhui Wan, Youyong Kong, Tao Feng, Kezhong Zhang Published: 2025 ABSTRACT Parkinson’s Disease (PD) with Rapid Eye Movement Sleep Behavior Disorder (RBD) occurred before (body-first) and after (brain-first) motor symptoms may exhibit different MRI features. We aimed to investigate the structural and functional MRI pattern differences between brain-first and body-first subtypes of PD. 23 body-first and 19 brain-first PD patients, along with 20 matched healthy controls (HC) were enrolled and underwent T1-weighted and resting-state functional magnetic resonance imaging (rs-fMRI) scans in NJ-dataset. We calculated and compared amplitude of low-frequency fluctuations (ALFF) and gray matter volume (GMV) across groups to identify differential brain regions, which were subsequently extracted in PPMI and OASIS3 datasets. These values were combined with clinical data for binary classification machine learning training (with PPMI including 22 body-first and 35 brain-first patients used for feature selection and OASIS3 including 5 body-first and 10 brain-first patients used for external validation) and correlated with clinical scales. The body-first group exhibited higher zALFF values in the parietal lobe and greater GMV in the frontal lobe, while showing lower zALFF and GMV values in the cerebellum and subcortical nuclei (caudate nucleus for zALFF; medulla, hippocampus, amygdala, and olfactory bulb for GMV) than brain-first group. We identified significant axial asymmetry in functional and structural MRI between brain-first and body-first Parkinson’s subtypes, characterized by greater gray matter retention and higher spontaneous neural activity in neocortex in the body-first subtype. Furthermore, integrating MRI and clinical scales effectively distinguished between brain-first and body-first subtypes. Introduction Parkinson’s Disease (PD) is a highly heterogeneous neurodegenerative disorder characterized by the degeneration of dopaminergic neurons in the substantia nigra. Notably, the accumulation of α-synuclein in the nigrostriatal system and other regions of the nervous system begins decades before the onset of motor symptoms1. Moreover, the burden associated with PD-related diseases continues to intensify annually, exacerbating the challenges faced by healthcare systems and the need for effective management strategies2–4. Rapid Eye Movement Sleep Behavior Disorder (RBD) is classified as an α-synucleinopathy, with over 80% of idiopathic RBD patients diagnosed with PD, Lewy Body Dementia, or Multiple System Atrophy at some point in their lives5,6. Given neuronal loss and Lewy pathology were found in pon of iRBD patients7,8, and the onset of parkinsonian motor symptoms is associated with midbrain damage, the timing of RBD onset relative to motor symptoms is hypothesized to reflect distinct spatiotemporal patterns of pathological progression, supporting its use as a subtyping criterion. Currently, the underlying mechanisms of PD remain unclear. The Braak staging hypothesis suggests that α-synuclein originates in the entorhinal cortex and subsequently spreads to the dorsal motor nucleus of the vagus and the substantia nigra, eventually accumulating in the neocortex, indicating a dispersive nature of α-synuclein9. Inspired by Braak’s hypothesis, researchers proposed in 2020 that PD can be classified into two subtypes: body-first and brain-first. The body-first (bottom-up) subtype is characterized by α-synuclein originating from autonomic and enteric nervous systems, progressing through the vagus and sympathetic nerves to the central nervous system. In contrast, the brain-first (top-down) subtype is defined by α-synuclein originating from central structures such as the olfactory bulb and limbic system, then transmitting to peripheral sites via the brainstem10. In clinical studies, to facilitate classification, some researchers define PD patients with RBD symptoms preceding motor symptoms as body-first, while those with RBD after motor symptoms are labeled brain-first11. Another approach categorizes patients with RBD symptoms at baseline as body-first and those without as brain-first12. Several data-driven studies on the progression of PD have found that body-first patients exhibit a faster progression of both motor and non-motor symptoms than brain-first patients12–15. Resting-state functional MRI studies have revealed more substantial disruptions in the locus coeruleus-related resting-state networks in body-first PD compared to brain-first PD16, with decreased functional connectivity between the right ventral posterior cingulate and the left medial precuneus, alongside increased normalized characteristic path length17. Structural imaging studies indicate that PD patients with RBD show higher cortical mean diffusivity in the left superior temporal gyrus, superior frontal gyrus, precentral gyrus, precuneus, and right middle frontal gyrus and postcentral gyrus compared to those without RBD. However, no significant differences in cortical thickness have been detected18. DaT SPECT studies suggest greater asymmetry in the brains of brain-first PD patients compared to body-first patients19, although findings from other studies reported contrasting results15,20. There is substantial evidence indicating that the brain-first and body-first PD subtypes can predict disease progression and prognosis among patients, with significant distinctions in disease course. However, whether there are distinct neuroimaging characteristics between these subtypes remains unresolved, as existing research has not reached consistent conclusions on this topic. This study aims to analyze the neuroimaging biomarkers that differentiate between brain-first and body-first PD subtypes and to utilize these findings, along with selected neuropsychological scales, as features in binary classification machine learning models. The goal is to identify the most valuable neuroimaging markers or combinations thereof based on classification efficacy. Results Demographic and clinical characteristics In NJ dataset, there were no significant differences in age, gender, education level, age at onset, Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Part III scores, Hoehn and Yahr stage, Rapid Eye Movement Sleep Behavior Disorder Screening Questionnaire (RBDSQ), and Mini-Mental State Examination (MMSE) scores among the three groups or between body-first and brain-first group. The body-first group exhibited significantly higher MDS-UPDRS-I scores (p = 0.005) than brain-first group. Additionally, compared to the HC group, body-first group displayed higher apathy scores, while the two PD groups showed significantly higher depression and anxiety scores (p < 0.05 after Bonferroni correction, see in Table 1). Demographic and clinical characteristics of the PPMI and OASIS3 cohort were shown in Supplementary Table 2 and Supplementary Table 4).Table 1Demographic and clinical characteristics of participants from NJ-datasetVariableBody-first (n = 23)Brain-first (n = 19)HC (n = 20)P-valuePost hoc (Bonferroni correction)Age (y)62.22 ± 8.1862.42 ± 6.7563.25 ± 6.040.885cGender (M/F)13/1011/812/80.974aEducation (y)10.65 ± 4.5011.68 ± 3.9411.75 ± 2.990.796bDisease duration (y)4.96 ± 3.477.53 ± 3.55NA0.023e,*Age at onset (y)57.48 ± 9.2355.00 ± 5.85NA0.317eMDS-UPDRSI5.65 ± 4.442.63 ± 1.74NA0.005d,**MDS-UPDRSIII26.04 ± 9.6028.42 ± 13.08NA0.501eHoehn & Yahr scale1.93 ± 0.382.12 ± 0.57NA0.225dRBDSQ8.35 ± 2.128.53 ± 2.39NA0.818dMMSE28.52 ± 1.4428.21 ± 1.3229.15 ± 0.930.069bHAMA13.00 ± 10.0412.32 ± 9.522.8 ± 3.49<0.001b,***body-first>HC (p = 0.002**) brain-first>HC (p = 0.002**)HAMD-2412.48 ± 12.0912.63 ± 10.882.3 ± 2.85<0.001b,***body-first>HC (p < 0.001***) brain-first>HC (p = 0.002**)AS20.04 ± 11.3613.68 ± 7.5610.65 ± 6.080.010b,*body-first>HC (p = 0.013*)Values except gender are presented as the mean ± standard deviation.aChi-square test.bKruskal-Wallis test.cA one-way analysis of variance (ANOVA) test.dMann-Whitney U test.eTwo-sample t-test.*p < 0.05, **p < 0.01, ***p < 0.001.RBD Rapid eye movement sleep behavior disorder, PD Parkinson’s Disease, body-first, RBD symptoms occurring before PD motor symptoms; brain-first: RBD symptoms occurring after PD motor symptoms, MDS-UPDRSI/III Movement Disorder Society Unified Parkinson’s Disease Rating Scale part Ⅰ/Ⅲ, RBDSQ RBD Screening Questionnaire, MMSE Mini-Mental State Examination, HAMA Hamilton Anxiety Rating Scale, HAMD-24 Hamilton Depression Rating Scale - 24 items, AS Apathy Scale. Table 1Demographic and clinical characteristics of participants from NJ-datasetVariableBody-first (n = 23)Brain-first (n = 19)HC (n = 20)P-valuePost hoc (Bonferroni correction)Age (y)62.22 ± 8.1862.42 ± 6.7563.25 ± 6.040.885cGender (M/F)13/1011/812/80.974aEducation (y)10.65 ± 4.5011.68 ± 3.9411.75 ± 2.990.796bDisease duration (y)4.96 ± 3.477.53 ± 3.55NA0.023e,*Age at onset (y)57.48 ± 9.2355.00 ± 5.85NA0.317eMDS-UPDRSI5.65 ± 4.442.63 ± 1.74NA0.005d,**MDS-UPDRSIII26.04 ± 9.6028.42 ± 13.08NA0.501eHoehn & Yahr scale1.93 ± 0.382.12 ± 0.57NA0.225dRBDSQ8.35 ± 2.128.53 ± 2.39NA0.818dMMSE28.52 ± 1.4428.21 ± 1.3229.15 ± 0.930.069bHAMA13.00 ± 10.0412.32 ± 9.522.8 ± 3.49<0.001b,***body-first>HC (p = 0.002**) brain-first>HC (p = 0.002**)HAMD-2412.48 ± 12.0912.63 ± 10.882.3 ± 2.85<0.001b,***body-first>HC (p < 0.001***) brain-first>HC (p = 0.002**)AS20.04 ± 11.3613.68 ± 7.5610.65 ± 6.080.010b,*body-first>HC (p = 0.013*)Values except gender are presented as the mean ± standard deviation.aChi-square test.bKruskal-Wallis test.cA one-way analysis of variance (ANOVA) test.dMann-Whitney U test.eTwo-sample t-test.*p < 0.05, **p < 0.01, ***p < 0.001.RBD Rapid eye movement sleep behavior disorder, PD Parkinson’s Disease, body-first, RBD symptoms occurring before PD motor symptoms; brain-first: RBD symptoms occurring after PD motor symptoms, MDS-UPDRSI/III Movement Disorder Society Unified Parkinson’s Disease Rating Scale part Ⅰ/Ⅲ, RBDSQ RBD Screening Questionnaire, MMSE Mini-Mental State Examination, HAMA Hamilton Anxiety Rating Scale, HAMD-24 Hamilton Depression Rating Scale - 24 items, AS Apathy Scale. Demographic and clinical characteristics of participants from NJ-dataset Values except gender are presented as the mean ± standard deviation. aChi-square test. bKruskal-Wallis test. cA one-way analysis of variance (ANOVA) test. dMann-Whitney U test. eTwo-sample t-test. *p < 0.05, **p < 0.01, ***p < 0.001. RBD Rapid eye movement sleep behavior disorder, PD Parkinson’s Disease, body-first, RBD symptoms occurring before PD motor symptoms; brain-first: RBD symptoms occurring after PD motor symptoms, MDS-UPDRSI/III Movement Disorder Society Unified Parkinson’s Disease Rating Scale part Ⅰ/Ⅲ, RBDSQ RBD Screening Questionnaire, MMSE Mini-Mental State Examination, HAMA Hamilton Anxiety Rating Scale, HAMD-24 Hamilton Depression Rating Scale - 24 items, AS Apathy Scale. Group difference in amplitude of low-frequency fluctuations (ALFF) In NJ database, body-first group exhibited higher zALFF values in the left inferior parietal lobe and lower zALFF values in the cerebellar vermis and left caudate nucleus compared to brain-first group. Both PD groups showed higher zALFF values in the inferior temporal gyrus, right inferior frontal triangle, and cerebellum, as well as lower zALFF values in the left cuneus compared to the HC group. (voxel p < 0.05 corrected by FDR, cluster p < 0.05 corrected by FWE, see in Table 2 and Fig. 1).Fig. 1ALFF comparisons among body-first, brain-first and HC in NJ-dataset.The statistical significance threshold was set at FDR corrected voxel-wise (p < 0.05) and FWE corrected cluster-wise (p < 0.05). Comparison between (A) body-first and brain-first, (B) body-first and HC, and (C) brain-first and HC. Red represents significantly higher zALFF and blue represents significantly lower zALFF values in the former group. body-first: RBD symptoms occurring before PD motor symptoms; brain-first: RBD symptoms occurring after PD motor symptoms, HC healthy control, FDR false discovery rate, FWE family-wise error, ALFF amplitude of low-frequency fluctuations.Table 2Comparisons of ALFF among body-first, brain-first and HC from NJ-datasetBrain regions(AAL3)Peak MNI CoordinatesNumber of voxelsPeak T valuePeak P value (FDR-corr)Cluster P-value (FWE-corr)xyzbody-first vs brain-firstParietal_Inf_L−42−5454384.850.0190.002Caudate_L−18−60234−5.100.0150.001Vermis_96−57−39127−4.000.0230.033body-first vs HCTemporal_Inf_L−54−18−30112345.6800Temporal_Inf_R51−12−3638238.8100.026Cerebellum_8_R21−63−4856912.0700.003Frontal_Inf_Tri_R6039155147.3500.006Cuneus_L0−90303695−22.5000brain-first vs HCTemporal_Inf_L−54−18−30109841.2800Temporal_Inf_R51−12−3639434.4800.022Cerebellum_8_R21−63−4856710.7700.003Frontal_Inf_Tri_R6039154685.3300.009Cuneus_L0−90303802−20.0400ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the ALFF. The statistical significance threshold was set at false discovery rate (FDR) corrected voxel-wise (p < 0.05) and family-wise error (FWE) corrected cluster-wise (p < 0.05).AAL3 anatomical automatic labeling atlas 3, ALFF Amplitude of Low-Frequency Fluctuations, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. ALFF comparisons among body-first, brain-first and HC in NJ-dataset. The statistical significance threshold was set at FDR corrected voxel-wise (p < 0.05) and FWE corrected cluster-wise (p < 0.05). Comparison between (A) body-first and brain-first, (B) body-first and HC, and (C) brain-first and HC. Red represents significantly higher zALFF and blue represents significantly lower zALFF values in the former group. body-first: RBD symptoms occurring before PD motor symptoms; brain-first: RBD symptoms occurring after PD motor symptoms, HC healthy control, FDR false discovery rate, FWE family-wise error, ALFF amplitude of low-frequency fluctuations. Table 2Comparisons of ALFF among body-first, brain-first and HC from NJ-datasetBrain regions(AAL3)Peak MNI CoordinatesNumber of voxelsPeak T valuePeak P value (FDR-corr)Cluster P-value (FWE-corr)xyzbody-first vs brain-firstParietal_Inf_L−42−5454384.850.0190.002Caudate_L−18−60234−5.100.0150.001Vermis_96−57−39127−4.000.0230.033body-first vs HCTemporal_Inf_L−54−18−30112345.6800Temporal_Inf_R51−12−3638238.8100.026Cerebellum_8_R21−63−4856912.0700.003Frontal_Inf_Tri_R6039155147.3500.006Cuneus_L0−90303695−22.5000brain-first vs HCTemporal_Inf_L−54−18−30109841.2800Temporal_Inf_R51−12−3639434.4800.022Cerebellum_8_R21−63−4856710.7700.003Frontal_Inf_Tri_R6039154685.3300.009Cuneus_L0−90303802−20.0400ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the ALFF. The statistical significance threshold was set at false discovery rate (FDR) corrected voxel-wise (p < 0.05) and family-wise error (FWE) corrected cluster-wise (p < 0.05).AAL3 anatomical automatic labeling atlas 3, ALFF Amplitude of Low-Frequency Fluctuations, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. Comparisons of ALFF among body-first, brain-first and HC from NJ-dataset ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the ALFF. The statistical significance threshold was set at false discovery rate (FDR) corrected voxel-wise (p < 0.05) and family-wise error (FWE) corrected cluster-wise (p < 0.05). AAL3 anatomical automatic labeling atlas 3, ALFF Amplitude of Low-Frequency Fluctuations, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. There was significant axial asymmetry in MRI functional characteristics between the brain-first (top-down) and body-first (bottom-up) groups. In body-first subtype, the neocortex exhibited higher spontaneous neural activity than cerebellum and caudate. Group Difference in Gray Matter Volume (GMV) In NJ database, body-first group exhibited higher GMV values in the right inferior frontal operculum and lower GMV values in the cerebellum, medulla, left precentral gyrus, left hippocampus, left amygdala, and right olfactory bulb compared to brain-first group. Both PD groups showed lower GMV values across nearly the entire brain compared to the HC group, except for higher GMV in the culmen for brain-first group relative to the HC group (voxel p < 0.001(unc), cluster p < 0.01 corrected by FDR, see in Table 3 and Fig. 2).Fig. 2GMV comparisons among body-first, brain-first and HC in NJ-dataset.The statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and FDR corrected cluster-wise (p < 0.01). Comparison between (A) body-first and brain-first, (B) body-first and HC, and (C) brain-first and HC. Red represents significantly higher GMV and blue represents significantly lower GMV values in the former group. body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control, FDR false discovery rate, unc uncorrection, GMV gray matter volume.Table 3Comparisons of GMV among body-first, brain-first and HC from NJ-datasetBrain regions(AAL3)Peak MNI CoordinatesNumber of voxelsPeak T valuePeak P-value (unc)Cluster P-value (FDR-corr)xyzbody-first vs brain-firstFrontal_Inf_Oper_R509145644.0500.007Cerebellum_8_R15−80−541300−4.8600Precentral_L−30−2670420−4.400.008Cerebellum_9_L, Medulla−6−34−52561−4.1300.002Vermis_6, Culmen8−63−121871−4.1300Hippocampus_L, Amygdala_L−3−4−181048−4.0800Olfactory_R28−2503−3.7700.003body-first vs HCPrecentral_R15−33751368−6.300Temporal_Mid_R63−42142027−6.0300Postcentral_L−15−26752362−5.7400Cerebellum_Crus2_L−24−81−523013−5.4100Temporal_Sup_L−40−8142034−5.3200Angular_R52−64281409−5.300Thal_VL_L−9−414573−5.300Precentral_R51−1257527−4.9900.001Insula_R56−231922−4.8800Calcarine_L−12−100−12561−4.7600Cerebellum_8_R16−84−442139−4.7600Insula_R−62−36−101299−4.7500Cerebellum_Crus1_L−33−78−181366−4.6400Cerebellum_6_R51−57−323650−4.6100Rectus_L022−91511−4.4500Temporal_Pole_Mid_L−15−4−12996−3.8500Frontal_Sup_2_R345721543−3.5900Parietal_Inf_L−38−4538416−3.9900.003ACC_pre_L25010414−3.6100.003brain-first vs HCCerebellum_4_5_L, Culmen−16−21−309998.700Temporal_Mid_R60−28123174−5.3100Precentral_R−9−3978482−5.2600.002Cerebellum_8_L−44−57−572088−5.0600Parietal_Inf_L−38−4639514−5.0500.001Temporal_Pole_Sup_L−244−333129−4.8500ACC_pre_R05110583−4.600Insula_R5414−21161−4.5200Occipital_Mid_R33−6948468−4.500.002Calcarine_R20−46−3841−4.400Temporal_Mid_L−58−54−3555−4.3100.001Cerebellum_Crus1_L−38−75−20489−4.300.001Fusiform_R32−12−33884−4.2100Postcentral_L−51−1230847−4.0200Cerebellum_Crus1_R45−63−22417−3.9900.004Frontal_Inf_Tri_R52429689−3.7200ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the GMV. The statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and false discovery rate (FDR) corrected cluster-wise (p < 0.01).AAL3 anatomical automatic labeling atlas 3, GMV Gray Matter Volume, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. GMV comparisons among body-first, brain-first and HC in NJ-dataset. The statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and FDR corrected cluster-wise (p < 0.01). Comparison between (A) body-first and brain-first, (B) body-first and HC, and (C) brain-first and HC. Red represents significantly higher GMV and blue represents significantly lower GMV values in the former group. body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control, FDR false discovery rate, unc uncorrection, GMV gray matter volume. Table 3Comparisons of GMV among body-first, brain-first and HC from NJ-datasetBrain regions(AAL3)Peak MNI CoordinatesNumber of voxelsPeak T valuePeak P-value (unc)Cluster P-value (FDR-corr)xyzbody-first vs brain-firstFrontal_Inf_Oper_R509145644.0500.007Cerebellum_8_R15−80−541300−4.8600Precentral_L−30−2670420−4.400.008Cerebellum_9_L, Medulla−6−34−52561−4.1300.002Vermis_6, Culmen8−63−121871−4.1300Hippocampus_L, Amygdala_L−3−4−181048−4.0800Olfactory_R28−2503−3.7700.003body-first vs HCPrecentral_R15−33751368−6.300Temporal_Mid_R63−42142027−6.0300Postcentral_L−15−26752362−5.7400Cerebellum_Crus2_L−24−81−523013−5.4100Temporal_Sup_L−40−8142034−5.3200Angular_R52−64281409−5.300Thal_VL_L−9−414573−5.300Precentral_R51−1257527−4.9900.001Insula_R56−231922−4.8800Calcarine_L−12−100−12561−4.7600Cerebellum_8_R16−84−442139−4.7600Insula_R−62−36−101299−4.7500Cerebellum_Crus1_L−33−78−181366−4.6400Cerebellum_6_R51−57−323650−4.6100Rectus_L022−91511−4.4500Temporal_Pole_Mid_L−15−4−12996−3.8500Frontal_Sup_2_R345721543−3.5900Parietal_Inf_L−38−4538416−3.9900.003ACC_pre_L25010414−3.6100.003brain-first vs HCCerebellum_4_5_L, Culmen−16−21−309998.700Temporal_Mid_R60−28123174−5.3100Precentral_R−9−3978482−5.2600.002Cerebellum_8_L−44−57−572088−5.0600Parietal_Inf_L−38−4639514−5.0500.001Temporal_Pole_Sup_L−244−333129−4.8500ACC_pre_R05110583−4.600Insula_R5414−21161−4.5200Occipital_Mid_R33−6948468−4.500.002Calcarine_R20−46−3841−4.400Temporal_Mid_L−58−54−3555−4.3100.001Cerebellum_Crus1_L−38−75−20489−4.300.001Fusiform_R32−12−33884−4.2100Postcentral_L−51−1230847−4.0200Cerebellum_Crus1_R45−63−22417−3.9900.004Frontal_Inf_Tri_R52429689−3.7200ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the GMV. The statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and false discovery rate (FDR) corrected cluster-wise (p < 0.01).AAL3 anatomical automatic labeling atlas 3, GMV Gray Matter Volume, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. Comparisons of GMV among body-first, brain-first and HC from NJ-dataset ANCOVA and post hoc two-sample t-tests were performed to examine between-groups differences of the GMV. The statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and false discovery rate (FDR) corrected cluster-wise (p < 0.01). AAL3 anatomical automatic labeling atlas 3, GMV Gray Matter Volume, L left, R right, MNI Montreal Neurological Institute, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, HC healthy control. The MRI structural characteristics showed significant axial asymmetry between the two groups. The neocortex exhibited greater gray matter preservation compared to subcortical and infratentorial structures in body-first group. Classification machine learning results Using imaging features alone yielded limited classification performance (best AUC = 0.63, accuracy = 0.77; permutation p = 0.009). Integrating imaging with clinical scales improved classification, as shown in Fig. 3.Fig. 3Performances of KNN Model with Different Feature Combinations.Performance of KNN Model for classification between body-first and brain-first with Different Combinations of (A–C) 10 Imaging Features and (D, E) 10 Imaging Features + 8 clinical features, implementing permutation test. (A, D) Relationship between AUC, p-value and Accuracy for KNN Model with different feature combinations; (B, C) Accuracy and (E, F) AUC distributions for KNN model with different feature combinations. KNN K-Nearest Neighbors, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, AUC area under the curve, p-value permutation p-value. Performances of KNN Model with Different Feature Combinations. Performance of KNN Model for classification between body-first and brain-first with Different Combinations of (A–C) 10 Imaging Features and (D, E) 10 Imaging Features + 8 clinical features, implementing permutation test. (A, D) Relationship between AUC, p-value and Accuracy for KNN Model with different feature combinations; (B, C) Accuracy and (E, F) AUC distributions for KNN model with different feature combinations. KNN K-Nearest Neighbors, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, AUC area under the curve, p-value permutation p-value. Two distinct performance thresholds were applied. The first threshold was met by 54 combinations (AUC & accuracy ≥0.80, sensitivity ≥0.70, specificity ≥0.70, p ≤ 0.05), and the second by 13 combinations (AUC & accuracy ≥0.80, sensitivity ≥0.65, specificity ≥0.75, p ≤ 0.01). The optimal feature set comprised nine high-performing features: lPrecentralGMV, MedullalCerebellumGMV, lParietalALFF, lCaudateALFF, VermisALFF, motor symptoms, RBDSQ, cognition symptoms and HY stage. Their occurrence in high-performance KNN combinations is detailed in Supplementary Table 5. Given the HY stage was not available in OASIS3, we systematically optimized and externally validated KNN and LDA models using all combinations of the remaining eight features. This process identified 2 high-performing combinations via KNN (AUC & accuracy ≥0.80 in PPMI and ≥0.75 in OASIS3; p ≤ 0.05), and 9 via LDA (AUC ≥ 0.80 & accuracy ≥0.75 in PPMI and AUC ≥ 0.75 & accuracy ≥0.70 in OASIS3; p ≤ 0.01), detailed in Fig. 4. The feature combination lCaudateALFF + VermisALFF + Motor symptoms demonstrated good performance across both the KNN and LDA models, detailed in Supplementary Fig. 2.Fig. 4Model performances of external validation with high-performance feature combinations.A, D Performance comparison between discovery dataset (PPMI) and external validation dataset (OASIS3); Relationship between AUC, p-value, and Accuracy in PPMI (B, E) and OASIS3 (C, F); (A–C) Using KNN; (D–F) Using LDA. KNN K-Nearest Neighbors, LDA Linear Discriminant Analysis, AUC area under the curve, p-value permutation p-value, Motor Movement Disorder Society Unified Parkinson’s Disease Rating Scale partⅢ in PPMI and UPDRS3 in OASIS3, Cognitio Montreal Cognitive Assessment in PPMI and Mini-Mental State Examination in OASIS3, ALFF amplitude of low-frequency fluctuations, lParietalALFF lCaudateALFF and VermisALFF mean zALFF value extracted from the cluster including the left inferior parietal lobe, left caudate or vermis with significant zALFF difference between body-first and brain-first groups in NJ-dataset, GMV gray matter volume, lPrecentralGMV and MedullalCerebellumGMV: GMV of the cluster including left precentral gyrus and the cerebellum, medulla with significant GMV difference between body-first and brain-first groups in NJ-dataset. Model performances of external validation with high-performance feature combinations. A, D Performance comparison between discovery dataset (PPMI) and external validation dataset (OASIS3); Relationship between AUC, p-value, and Accuracy in PPMI (B, E) and OASIS3 (C, F); (A–C) Using KNN; (D–F) Using LDA. KNN K-Nearest Neighbors, LDA Linear Discriminant Analysis, AUC area under the curve, p-value permutation p-value, Motor Movement Disorder Society Unified Parkinson’s Disease Rating Scale partⅢ in PPMI and UPDRS3 in OASIS3, Cognitio Montreal Cognitive Assessment in PPMI and Mini-Mental State Examination in OASIS3, ALFF amplitude of low-frequency fluctuations, lParietalALFF lCaudateALFF and VermisALFF mean zALFF value extracted from the cluster including the left inferior parietal lobe, left caudate or vermis with significant zALFF difference between body-first and brain-first groups in NJ-dataset, GMV gray matter volume, lPrecentralGMV and MedullalCerebellumGMV: GMV of the cluster including left precentral gyrus and the cerebellum, medulla with significant GMV difference between body-first and brain-first groups in NJ-dataset. Association between neuroimaging features and neuropsychological characteristics In NJ database, we did not find any imaging features associated with clinical characteristics that remained significant after multiple comparison corrections. However, we did identify a near-significant negative correlation between the GMV of the left limbic system and MDS-UPDRS I (r = −0.43, p = 0.048, unc) in body-first group. In contrast, no correlations were found in brain-first group (r = 0.044, p = 0.86, unc, see in Fig. 5). This indicates that the association between non-motor symptoms and imaging changes may exist only in the body-first subtype.Fig. 5Pearson’s correlation between imaging features and clinical features in NJ-dataset.Pearson’s correlation between lLimbicGMV and MDS-UPDRS-Ⅰ in (A) all PD patients, (B) body-first patients, and (C) brain-first patients in NJ dataset, the p values were uncorrected. MDS-UPDRS-Ⅰ The Movement Disorder Society Unified Parkinson’s Disease Rating Scale Part I, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, RBD Rapid Eye Movement Sleep Behavior Disorder, GMV gray matter volume, lLimbicGMV the GMV of the cluster including left limbic systems with significant GMV difference between body-first and brain-first. Pearson’s correlation between imaging features and clinical features in NJ-dataset. Pearson’s correlation between lLimbicGMV and MDS-UPDRS-Ⅰ in (A) all PD patients, (B) body-first patients, and (C) brain-first patients in NJ dataset, the p values were uncorrected. MDS-UPDRS-Ⅰ The Movement Disorder Society Unified Parkinson’s Disease Rating Scale Part I, body-first RBD symptoms occurring before PD motor symptoms, brain-first RBD symptoms occurring after PD motor symptoms, RBD Rapid Eye Movement Sleep Behavior Disorder, GMV gray matter volume, lLimbicGMV the GMV of the cluster including left limbic systems with significant GMV difference between body-first and brain-first. Discussion Our study found a potential neuroimaging axial asymmetry between brain-first (top-down) and body-first (bottom-up) subtypes in both functional and structural MRI, characterized by greater gray matter retention and higher spontaneous neural activity in neocortex in body-first subtype, see in Fig. 6. Axial MRI asymmetry in PD patients with RBD may serve as a potential neuroimaging marker for distinguishing between the two subtypes. Moreover, our research further demonstrated through a systematic machine learning binary classification approach that neuroimaging features combined with clinical scales can effectively distinguish between the two subtypes.Fig. 6Schematic Illustrating the Axial Asymmetry in body-first and brain-first PD.A Body-first PD: Characterized by greater gray matter retention and higher spontaneous neural activity in the neocortex. B Brain-first PD: Characterized by reduced gray matter volume and lower spontaneous neural activity in the neocortex. Created in BioRender. Shi, J. (2025) https://BioRender.com/ltgy3aq. Schematic Illustrating the Axial Asymmetry in body-first and brain-first PD. A Body-first PD: Characterized by greater gray matter retention and higher spontaneous neural activity in the neocortex. B Brain-first PD: Characterized by reduced gray matter volume and lower spontaneous neural activity in the neocortex. Created in BioRender. Shi, J. (2025) https://BioRender.com/ltgy3aq. We found greater gray matter loss and lower spontaneous neural activity in infratentorial (medulla and cerebellum) and subcortical structures (hippocampus, amygdala, olfactory bulb, and caudate nucleus), which aligns with the brain-first versus body-first hypothesis. According to this hypothesis21–25, brain-first patients would be expected to exhibit more severe amygdala and olfactory bulb damage than body-first patients. However, our findings contradict this expectation. Several factors can explain this discrepancy. First, gray matter atrophy in PD resulted from a combination of Lewy body pathology, selective neuronal vulnerability26, and iron deposition27, as neuronal loss does not always align with Lewy body distribution28. Second, the Synuclein Origin and Connectome (SOC) model29 proposed that α-synuclein accumulation originates stochastically. While it often begins in the gut or olfactory bulb/amygdala, multifocal onset occurs in a minority (~10%) of patients30,31. Third, recent data-driven studies have revealed heterogeneous spatiotemporal trajectories of gray matter atrophy in PD. For instance, one study32 identified three distinct atrophy subtypes with origins in neocortical, limbic, and brainstem, while another33 established “cortex-first” and “deep gray-first” subtypes. Collectively, these findings demonstrated the existence of multiple, distinct patterns of structural degeneration in PD. Furthermore, an additional study27 found that cortical iron deposition may precede the limbic system in PD with higher dementia risk. While our results suggested potential neuroimaging biomarkers, the underlying neuropathological mechanisms require further exploration. The best feature combination for distinguishing body-first from brain-first PD comprised ALFF in the caudate and vermis combined with motor symptoms. This is consistent with the well-established faster motor progression in body-first PD34,35. An epicenter modeling study36 identified two subtypes: one with epicenters in cerebellar and midbrain, and the other with cortical and striatal. The former was associated with severe motor symptoms and rapid progression. Cerebellum was vulnerable to pathological α-synuclein accumulation and is increasingly implicated in modulating both motor and non-motor circuits in PD37,38. Our findings thus provide further evidence supporting the role of the cerebellum in structural degeneration within the body-first subtype. Moreover, our study identified two subtypes of axial asymmetry rather than left-right asymmetry. The brain-first and body-first hypothesis posits that the progression of PD from different origins is asymmetric19, a claim that remains a topic of debate15,20,39. Past studies have focused on the asymmetries between the left and right hemispheres of the brain. They suggest that pathological proteins transmitted via the vagus nerve spread bilaterally, while proteins originating in the brain are less likely to disseminate to both sides. Despite its significance, axial asymmetry has been under-investigated. Our findings could enhance understanding of structural progression in PD and suggest candidate therapeutic targets. It would be valuable to investigate in future studies whether body-first PD patients, characterized by infratentorial alterations, are more responsive to cerebellar neuromodulation than brain-first patients. Classifying PD patients with concurrent RBD into “brain-first” and “body-first” subtypes yielded improved classification performance. PD exhibits considerable heterogeneity, making it challenging to categorize all patients into simple binary subtypes. This can explain the inconsistencies found in previous studies on brain-first versus body-first. Previous research has shown that only a small subset of patients initially presenting with RBD experiences its later disappearance40, so the NJ dataset demonstrated lower patient heterogeneity within each group when excluding those who did not experience RBD throughout their disease course. Exploring brain-first and body-first subtypes only among patients who have already developed RBD is a more effective research approach. Furthermore, we found that body-first patients exhibit higher levels of non-motor symptoms compared to brain-first patients, and these non-motor symptoms were near-significantly correlated with the GMV of the left limbic system. However, such associations are not present in the brain-first subtype. It is well established across multiple studies that body-first patients experience more rapidly progressing non-motor symptoms13,14,41,42. Previous studies indicate that amygdala damage significantly affects both healthy individuals and PD patients43,44. We propose that limbic GMV loss in body-first patients was not merely a correlate but a driver of non-motor symptoms. Supporting this, a “subcortex-first” iRBD subtype with limbic-onset atrophy has been identified45, and subcortical structural deficits are seen in iRBD patients with cognitive impairment46. A subset of body-first PD patients may already exhibit limbic atrophy at the iRBD stage. In contrast, non-motor symptoms in brain-first PD likely originate from alternative mechanisms. Our study had several limitations. First, the modest sample size limited the statistical power of our neuroimaging analyses and increased the risk of overfitting in machine learning models. These limitations necessitate cautious interpretation of the results, due to the risks of both false positives and false negatives, and underscore the need for validation in larger longitudinal cohorts. Second, the use of different multiple comparison correction thresholds for ALFF (FDR and FWE) and GMV (voxel-level uncorrected p < 0.001 with FDR at the cluster level) precludes direct statistical comparisons between the two modalities. Therefore, the relationship between GMV and ALFF should be interpreted with caution. Third, for external validation, we utilized the OASIS3 database, which lacks information on the RBDSQ, HY stage, and other clinical scales. This omission prevented the evaluation of these clinical correlates. Finally, the diagnosis of RBD in our study was not confirmed by polysomnography (PSG). Future studies should seek to validate our findings using PSG-confirmed RBD diagnoses. In Summary, we found significant axial asymmetry between brain-first (top-down) and body-first (bottom-up) subtypes in both functional and structural MRI. We further demonstrated through a machine learning binary classification approach that the functional and structural MRI, combined with clinical scales, can effectively distinguish between the two subtypes. Methods Participants Data for this study were sourced from three databases. The NJ database includes patients admitted to the Neurology Department of Jiangsu Provincial People’s Hospital from September 2022 to April 2024, comprising 62 participants. Among them, there are 23 body-first patients, where RBD symptoms occurred before the onset of Parkinson’s motor symptoms and 19 brain-first patients, where RBD symptoms occurred after the motor symptoms of PD, and 20 age- and sex-matched healthy controls (HC). Diagnosis of PD was validated by two specialists in movement disorders (KZ Zhang and YS Yuan) based on the Movement Disorder Society clinical diagnostic criteria47. All participants were free from significant and acute medical conditions, neurological disorders, affective disorders, psychotic illnesses, or cognitive impairments, as indicated by a Mini-Mental State Examination (MMSE) score ≥24. Furthermore, individuals with an indeterminate temporal sequence between RBD and Parkinson’s motor symptoms due to ambiguous recall or lack of a bedpartner were also excluded. Fifty-seven patients were selected from the Parkinson’s Progression Markers Initiative (PPMI) (https://www.ppmi-info.org), including 35 brain-first and 22 body-first patients. Inclusion required available T1-weighted and fMRI images, data from ≥2 visits (covering diagnosis, RBDSSQ, MDS-UPDRS parts I/III, and neuropsychiatric scores), and a clear RBD-PD onset sequence. Exclusion criteria were: atypical parkinsonism; MoCA score <1848,49; or structural MRI indicating significant unrelated neurological pathology. The selection process is detailed in Supplementary Fig. 1, and subject IDs are listed in Supplementary Table 1. Fifteen patients were included from the Open Access Series of Imaging Studies3 (OASIS3) (https://sites.wustl.edu/oasisbrains/), comprising 10 brain-first and 5 body-first patients. Similar inclusion/exclusion criteria to PPMI were applied. See Supplementary Table 3 for included subject IDs. The entire protocol was approved by the ethics committee of the First Affiliated Hospital of Nanjing Medical University. Every PPMI recruitment center involved in this study obtained ethical approval from its respective local ethics committee. Informed consent was obtained from all OASIS3 participants under a protocol approved by the Washington University Institutional Review Board (IRB). All participants provided written informed consent by the principles of the Declaration of Helsinki. Clinical evaluation All participants underwent comprehensive neurological and neuropsychological assessments. The RBDSQ was utilized to evaluate REM sleep behavior disorder5. Probable RBD was diagnosed based on an RBDSQ threshold of ≥6. In addition, comprehensive interviews were conducted with all participants and their bed partners concerning the emergence of RBD symptoms and Parkinsonism50. The MDS-UPDRS Part I assesses the non-motor aspects of daily living, including cognitive impairment, emotional disturbances, and sleep issues. MDS-UPDRS Part III evaluates motor symptoms of PD through a series of clinical examinations, including tremor, rigidity, bradykinesia, and postural instability. The Hoehn & Yahr scale categorizes the severity of PD into five stages, aiding in the determination of disease progression. The MMSE and MoCA evaluate various cognitive domains, including orientation, memory, attention, and language skills. The Hamilton Anxiety Rating Scale (HAMA) and State-Trait Anxiety Inventory (STA) assess the severity of anxiety symptoms, while the Hamilton Depression Rating Scale (HAMD-24) and Geriatric Depression Scale (GDS) measure the severity of depression. The Apathy Scale (AS) and MDS-UPDRS I, Question 5 evaluates levels of apathy. Large cohort studies tracking non-motor symptoms in PD patients with and without RBD have revealed that PD patients with RBD exhibit significantly higher depression scores and an increased risk of mild cognitive impairment42. This indicates that our study’s assessment of non-motor symptoms focused on emotional and cognitive aspects. Clinical scales were conducted on the same day as the imaging scans. Given the availability of only binary RBD status (present/absent) in OASIS3, the RBDSQ scores in PPMI were converted into a corresponding binary variable for the cross-dataset validation analysis. MRI acquisition In the NJ dataset, MRI scans were conducted using a 3.0 T Siemens MAGNETOM Verio whole-body MRI system (Siemens Medical Solutions, Germany) equipped with an eight-channel phased-array head coil. We utilized snug foam padding to minimize head movement and provided earplugs to reduce noise. Participants were instructed to remain still, avoid any thoughts, and refrain from falling asleep without our reminders. The scans employed a three-dimensional magnetization-prepared rapid gradient echo (MP-RAGE) sequence with the following parameters: repetition time (TR) = 1900 ms, echo time (TE) = 2.95 ms, flip angle (FA) = 9°, slice thickness = 1 mm, number of slices = 160, field of view (FOV) = 230 × 230 mm2, matrix size = 256 × 256, and voxel size = 1 × 1 × 1 mm3. Additionally, echo-planar imaging (EPI) was used with the following parameters: TR = 2000 ms, TE = 21 ms, FA = 90, FOV = 256 × 256 mm2, in-plane matrix = 64 × 64, number of slices = 35, slice thickness = 3 mm (no gap), voxel size = 3 × 3 × 3 mm3, total volumes = 240. For PPMI dataset, data were acquired across multiple sites using 3-Tesla scanners (Siemens, Philips, GE) and a 1.5-Tesla scanner (Toshiba). The parameters for T1-weighted imaging varied as follows: TR = 2300 ms for Siemens scanners and 4.0–7.0 ms for others; TE = 1.92–3.06 ms for Siemens and 1.83–2.99 ms for others; FA = 9/15°; slice thickness = 1.0 mm; and matrix size = 256 × 256. The resting-state fMRI parameters were: TR = 1000/2500 ms; TE = 40 ms for Toshiba and 30 ms for others; FA = 58/80°; number of slices = 40–52; slice thickness = 5.0 mm for Toshiba and 3.0/3.5 mm for others; in-plane matrix = 64 × 64; and voxel size = 3.5 × 3.5 × 5.0 mm3 for Toshiba and 3.5 × 3.5 × 3.0–3.5 mm3 for others. For OASIS3 dataset, scans were predominantly performed on 3-Tesla Siemens TrioTim scanners. The T1-weighted imaging parameters were: TR = 2300/2400 ms; TE = 2.95–3.16 ms; FA = 8–9°; slice thickness = 1.0/1.2 mm; and matrix size = 256 × 256. The resting-state fMRI parameters were: TR = 2200–2500 ms; TE = 27 ms; FA = 90°; number of slices = 36; slice thickness = 4.0 mm; in-plane matrix = 64 × 64; and voxel size = 3.5 × 3.5 × 4.0 mm3. Processing of rs-fMRI data and ALFF analysis Preprocessing and analysis of functional images were conducted using the DPARSF software on the MATLAB R2021a platform51 (http://www.restfmri.net/forum/dparsf). The initial 10 time points were removed, followed by slice timing and head motion correction (Friston 24). Subjects exhibiting head motion exceeding 2.5 mm of translation or 2.5° of rotation in any direction were excluded from the analysis. The mean framewise displacement (FD) was computed, revealing no significant differences between the two groups. Skull stripping was applied to ensure precise co-registration of individual T1 structural and functional images. Additionally, signals from white matter, cerebrospinal fluid, and six head motion parameters were regressed out. High-pass temporal filtering (0.01–0.08 Hz) was performed to eliminate low-frequency drift and high-frequency physiological noise. The filtered images were then normalized to Montreal Neurological Institute (MNI) standard space using DARTEL technology (resampling voxel size = 3 × 3 × 3 mm3) and smoothed with a 6 mm full-width at half maximum (FWHM) Gaussian kernel. Any linear trends were subsequently removed. The calculation of ALFF (amplitude of low-frequency fluctuations) proceeded as follows: (1) All voxels were transformed from the time domain to the frequency domain using fast Fourier transform (FFT); (2) ALFF for each voxel was computed by averaging the square root of the power spectrum between 0.01 Hz and 0.08 Hz; (3) To standardize the results, the obtained ALFF values were converted to z-scores by subtracting the mean and dividing by the global standard deviation. Finally, the GCBH (GUI for ComBat Harmonization) tool on the MATLAB R2021a platform was used to harmonize the data from PPMI and OASIS352. Intergroup comparisons of zALFF from the NJ database were conducted to identify brain regions with significant differences between body-first and brain-first groups. Subsequently, the zALFF values were extracted from these identified clusters, and the average zALFF values in each cluster were calculated. Processing of T1-weighted images and GMV analysis Under MATLAB R2021a, T1-weighted images were preprocessed using version 12.7 of the Computational Anatomy Toolbox (CAT12) from SPM12 (https://www.fil.ion.ucl.ac.uk/spm; www.neuro.uni-jena.de/cat). The images were segmented into gray matter, white matter, and cerebrospinal fluid (CSF), and intensity non-uniformity was corrected. Spatial alignment was achieved through the DARTEL algorithm, which employs differential homomorphic anatomical registration to align data within the Montreal Neurological Institute (MNI) space. This process generated an average brain template for all participants before normalization in the MNI framework. Subsequently, gray matter images were modulated, and all resulting images were visually inspected to ensure quality. Following this, we used the GCBH toolbox to harmonize the gray matter images obtained from PPMI and OASIS3. The total intracranial volume (TIV) was estimated using the module available in CAT12. Intergroup comparisons of gray matter volume (GMV) images from the NJ database were conducted to identify brain regions with significant differences between body-first and brain-first groups. Thereafter, the GMV values were retrieved from these selected clusters. A linear regression model (residual method) was applied to regress total brain GMV against the TIV, yielding adjusted total brain GMV. Subsequently, we regressed each participant’s adjusted total brain GMV against the GMV of each cluster to obtain adjusted cluster GMV53. Machine learning model All analyses were implemented in MATLAB R2021a. The analytical framework incorporated ten imaging features, including three zALFF and seven GMV values extracted from significant clusters, and eight clinical measures: Hoehn & Yahr stage, RBDSQ, MDS-UPDRS-III, MDS-UPDRS-I, and recognition, depression, anxiety, and apathy scores. The machine learning pipeline followed a sequential validation approach, with all steps incorporating permutation testing (1000 iterations) to establish statistical significance. First, we conducted exhaustive feature screening. All possible combinations of both the ten imaging features and the full set of 18 integrated imaging-clinical features were systematically evaluated using a k-nearest neighbors (KNN) classifier within the PPMI dataset. This initial screening identified feature combinations meeting predefined performance thresholds (AUC and accuracy ≥0.80, sensitivity and specificity ≥0.7, permutation p < 0.05). The constituent features of these validated combinations were designated as high-performing for subsequent analysis. Next, we performed further feature combination analysis. All possible combinations of the identified high-performing features were evaluated using both KNN and Linear Discriminant Analysis (LDA) classifiers. During this phase, the PPMI dataset was partitioned into training (70%) and internal testing (30%) subsets. Model parameters were optimized through 5-fold cross-validation, and the resulting models were externally validated using the independent OASIS3 cohort. Finally, the generalizability of the best feature combination was assessed. The selected combination was evaluated on the independent OASIS3 dataset using both KNN and LDA classifiers. To mitigate potential overfitting given the limited sample size, model stability was further examined using 5-fold cross-validation. Statistical analysis Demographic and clinical characteristics of all subjects were compared by Chi-square test, two-sample t-test, Mann-Whitney U test, one-way analysis of variance (ANOVA), or Kruskal-Wallis test, as appropriate. R Statistical analyses (version 4.4.1) were employed, and p < 0.05 was considered statistically significant. Age, sex, education, MMSE scores, and mean FD were included as covariates in the analysis of covariance (ANCOVA) to compare zALFF values across the three groups. For post-hoc analysis, we used a two-sample t-test. The statistical significance threshold was set at false discovery rate (FDR) corrected voxel-wise (p < 0.05) and family-wise error (FWE) corrected cluster-wise (p < 0.05). Similar statistical analyses were conducted for GMV, where the covariate FD was replaced by TIV, and the statistical significance threshold was set at uncorrected voxel-wise (p < 0.001) and false discovery rate (FDR) corrected cluster-wise (p < 0.01). To minimize the false-positive rates of functional imaging results, we applied a rigorous correction at the cluster level. For structural imaging, we conducted exploratory analyses, thus using a more lenient method for multiple comparison correction. Subsequently, correlation was performed between the imaging features and the clinical scales that showed significant differences between body-first and brain-first groups, using Pearson correlation (p < 0.05), followed by FDR correction using the Benjamini-Hochberg method. Supplementary information Supplementary Materials Supplementary Materials