HBM-Shang Huifang's research group used functional imaging technology to reveal markers of motor progress in Parkinson's disease

Parkinson's disease (PD) is one of the most common neurodegenerative diseases [1].

Patients with PD develop a range of motor and non-motor symptoms, which have a significant negative impact on their quality of life and increase their socioeconomic burden
.
In addition, disease progression in PD is associated with increased disability and mortality [2].

Therefore, reliable monitoring of PD progression is essential to develop effective interventions, and its predictors will also help in the early identification of patients with rapidly progressing PD that require attention
.
However, reliable biomarkers of PD progression are currently unknown
.

Magnetic resonance imaging (MRI) can be used to objectively measure changes in brain structure and function, and therefore help identify biomarkers of PD progression
.
Resting-state functional MRI (rs-fMRI) can reveal functional reorganization of neural pathways before neuronal death or brain atrophy occurs in the brain.
Suggests the potential for early sensitivity [3].

Among them, Fractional amplitude of low-frequency fluctuation (fALFF) analysis can be used to determine the intensity of spontaneous brain activity in each brain region of the neural pathway [4].

。 A two-year longitudinal rs-fMRI study found that cerebellar fALFF values correlated with the Unified PD rating scale UPDRS) Part III score and score change were positively correlated, suggesting that the cerebellum may play an important role in PD motor progression [5]; Another study combined machine learning and fALFF analysis to propose a prediction algorithm that can establish UPDRS scores at future time points and capture robust statistical patterns [6].
These findings suggest that fALFF values provide sufficient information about neurobiological changes and are of great value
in prognostic prediction studies of PD.

On October 17, 2022, Shang Huifang's research group from the Department of Neurology, West China Hospital of Sichuan University, published a speech entitled in Human Brain Mapping (HBM).
"Motor progression marker for newly-diagnosed drug-naïve patients with Parkinson's disease: A resting-state functional MRI study" The baseline resting functional feature (i.
e.
, fALFF) is proposed to predict PD motor progression, and the caudate nucleus is the core brain region responsible for PD motor progression, which will help PD in clinical practice Development
of strategies to prevent disease progression.

By comparing whole-brain fALFF between the PD group and the healthy control (HC) group, the researchers found PD The fALFF values of the group were significantly reduced in the anterior central gyrus, posterior central gyrus, wedge lobe, lingual gyrus, and thalamus, while fALF values were significantly increased in the superior frontal gyrus and middle frontal gyrus (Figure 1A).

。 In addition, reduced function of the anterior and posterior central gyrus was associated with more severe motor symptoms (UPDRS Part III score) at baseline (Figure 1B).

Figure 1(A) Whole brain fALFF comparison results between PD group and HC group; (B) There was a significant negative correlation
between fALFF values in bilateral sensorimotor zones and baseline UPDRS Part III motor scores in all PD patients.

(Source: Hou, et al.
, HBM, 2022).

The researchers further divided the brain into 116 cortical and subcortical regions of interest (ROI) through automated anatomical marker atlases [7], and extracted them for each subject The fALFF value
of the ROI.
In order to obtain the most critical image features to distinguish PD and HC, a minimum absolute shrinkage and selection operator was performed LASSO) regression analysis
.
LASSO regression is a regression analysis method that can be used to identify variables, estimate the corresponding regression coefficients, and build models
with minimal error.
This method can be used to address overfitting of variables and the performance of the model too high in explaining variability
.

It was found that a classification model could be constructed based on the image features of nine brain regions, including the superior frontal gyrus, mid-frontal gyrus, anterior central gyrus, posterior central gyrus, inferior temporal gyrus, inferior cerebellar lobe, and putamen (Fig.
2A and 2B), and the coefficients of each variable in the model are shown in Fig.
2C shown
.
Notably, the image features of the right shell nucleus and the anterior central gyrus show the highest coefficients
.
The area under the curve (AUC) of this model is 0.
897 (95% CI: 0.
828-0.
967), sensitivity and specificity of 75.
0% and 92.
6%, respectively, with a cut-off of 0.
65 (Figure 2D ）
。 The ability of the model was tested using random sampling of internal datasets, i.
e.
, random sampling 47 times with sample sizes of 28 to 74, randomly drawing the test set from the entire cohort, and subsequently evaluating the AUC in different random samples
。 As shown in Figure 2E, all AUCs fluctuate around the model AUC, demonstrating good predictive power
.
The calibration curve shows good probabilistic agreement (Figure 2F), Hosmer-Lemeshow (HL The goodness-of-fit test showed no significant difference between observed and predicted events (p=0.
886).

Figure 2 Through LASSO regression analysis, the image features of 9 brain regions can be classified as a classification model

(Source: Hou, et al.
, HBM, 2022).

Using the same methods as above, the researchers found that predictive models could be constructed based on the image features of 12 brain regions, including the inferior frontal gyrus, olfactory cortex, hippocampus, temporal pole, superior occipital gyrus, upper and lower cerebellar lobes, thalamus, caudate nucleus, and putamen (Figures 3A and 3B).
The coefficients for each variable in the model are shown in
Figure 3C.
It is also worth noting that the image features of the right superior occipital gyrus and bilateral caudate nuclei show the highest coefficients
.
The AUC for this model was 0.
916 (95% CI: 0.
834-0.
999), with sensitivity and specificity of 100.
0%, respectively.
and 87.
0% with a cut-off of 0.
50 (Figure 3D).

The ability of the model was tested using random sampling of internal datasets, i.
e.
, 24 random samples with sample sizes of 24 to 47, as shown in Figure 3E The model has demonstrated good predictive power
.
The calibration curve also shows good probabilistic agreement (Figure 3F), again the HL test did not produce significant deviations between observed and predicted events ().
p=0.
198）
。

Figure 3 Through LASSO regression analysis, the image features of 12 brain regions can build predictive models

(Source: Hou, et al.
, HBM, 2022).

Comparing the salient imaging features of the above model, the researchers found that relative to the HC group, the fALFF values in the putamen and sensorimotor regions were significantly reduced in the PD group, and fALFF in the frontal and temporal lobes Values are significantly elevated (Figure 4A).

In addition, fALFF values were significantly lower in the supraoccipital gyrus of the PD subgroup with slow motor progression compared to the rapidly progressive PD subgroup, but fALFF in the caudate nucleus, hippocampus, and inferior frontal gyrus The value increases significantly (Figure 4B).

In addition, the results of partial correlation analysis found a significant negative correlation between fALFF values of bilateral caudate nuclei and annual increase scores of UPDRS Part III scores in the PD group (right caudate nucleus: r=- 0.
345； p=0.
019； Left caudate nucleus; r=-0.
34； p=0.
0019) (Figure 4C).

Figure 4 Post-hoc analysis of salient image features of the model

(Source: Hou, et al.
, HBM, 2022).