Detect the disrupted brain structural connectivity in type 2 diabetes mellitus patients without cognitive impairment

Table 1 Demographics and clinical data of type 2 diabetes mellitus and healthy controls

	T2DM subjects, (n = 47)	Healthy controls, (n = 47)	P value
Demographics
Age (years)	57.36 ± 8.04	55.62 ± 8.36	0.305
Gender (male/female)	18/29	25/22	0.147
Clinical data
BMI (kg/m2)	24.52 ± 2.09	20.78 ± 1.40	0.0001
SBP (mmHg)	124.15 ± 9.66	114.51 ± 6.67	0.0001
DBP (mmHg)	75.62 ± 6.54	70.36 ± 5.14	0.0001
HbA1c (%)	8.03 ± 1.53	N/A	N/A
FBG (mmol/L)	6.73 ± 2.15	N/A	N/A
TG (mmol/L)	1.65 ± 0.42	N/A	N/A
TC (mmol/L)	4.93 ± 1.03	N/A	N/A
HDL (mmol/L)	1.00 ± 0.20	N/A	N/A
LDL (mmol/L)	2.59 ± 0.53	N/A	N/A
Course of disease (years)	6.60 ± 3.41	N/A	N/A

¹P < 0.05.

Values distributed normally or non-normally are presented as mean ± SD or median (minimum, maximum). HbA1c: Hemoglobin A1C; BMI: Body mass index; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; FBG: Fasting blood glucose; TG: Triglyceride; TC: Total cholesterol; LDL: Low density lipoprotein; NA: Not Applicable.

Table 2 Performance Evaluation for type 2 diabetes mellitus classification models

Models	Accuracy	Precision	Recall	F1-score
Connectivity baseline model	0.5011 (0.0064)	0.5009 (0.0085)	0.5333 (0.0195)	0.5104 (0.0105)
Local surface model	0.4678 (0.0056)	0.4840 (0.0103)	0.4733 (0.0161)	0.4649 (0.0056)
Local volume model	0.3614 (0.0057)	0.3478 (0.0111)	0.3844 (0.0274)	0.3610 (0.0179)
Local baseline model	0.3824 (0.0066)	0.3616 (0.0129)	0.4044 (0.0376)	0.3772 (0.0226)
Advanced connectivity models
Logistic	0.8099 (0.0071)	0.7983 (0.0072)	0.8311 0.0162)	0.8104 (0.0083)
Ridge (α = 0.9)	0.8304 (0.0058)	0.8694 (0.0066)	0.7867 (0.0189)	0.8185 (0.0089)
SVM	0.8093 (0.0026)	0.8127 (0.0035)	0.8088 (0.0106)	0.8065 (0.0038)
Connectivity interaction models
Logistic	0.8204 (0.0071)	0.8011 (0.0076)	0.8511 (0.0125)	0.8232 (0.0081)
Ridge (α = 0.2)	0.9251 (0.0007)	0.9377 (0.0026)	0.9155 (0.0018)	0.9249 (0.0006)
SVM	0.8204 (0.0049	0.8350 (0.0089)	0.8088 (0.00106)	0.8170 (0.0059)

SVM: Supporting vector machine; Logistic: Logistic classification; Ridge: Ridge classification (α is the hyper-parameter α tuning selected by the model). The process of the connection feature and interaction term selection, as well as the model evaluation metrics were included in the Supplementary material.

Table 3 The 18 selected connectivity features for type 2 diabetes mellitus classification models

Connections	Ridge regression coefficients
Parietal_Inf_R-SupraMarginal_R	3.3436
Supp_Motor_Area_R-Cingulum_Mid_R	3.1377
Cuneus_L-Caudate_L	2.0824
Frontal_Sup_Orb_L-Parietal_Inf_L	1.7015
Supp_Motor_Area_L-Supp_Motor_Area_R	1.5698
Frontal_Mid_Orb_R-Thalamus_R	1.5043
Calcarine_R-Thalamus_R	1.3564
Supp_Motor_Area_R-Paracentral_Lobule_L	1.3261
Frontal_Med_Orb_R-Cingulum_Mid_R	1.1464
Parietal_Inf_R-Temporal_Inf_R	0.9017
ParaHippocampal_R-Occipital_Sup_R	0.8107
Frontal_Med_Orb_L-Pallidum_L	0.8031
Calcarine_L-Fusiform_L	0.7449
Caudate_L-Pallidum_L	0.6747
Frontal_Sup_Medial_R-Putamen_R	0.5617
Frontal_Med_Orb_L-Putamen_L	0.5418
Fusiform_R-Thalamus_R	0.4652
Cingulum_Post_R-Occipital_Sup_L	0.3565

Please refer to the AAL-116 template for the names and locations of the brain regions of interest listed in the table. The table displays the absolute values of the coefficients for each feature (sorted from highest to lowest) in the model with the best performance, the Ridge regression model. A larger coefficient value indicates a greater contribution of the feature to the prediction outcome. The connections associated with the deep subcortical nuclei were highlighted in green.

Table 4 Performance evaluation for type 2 diabetes mellitus disease duration prediction models

Models	RMSE	Var (RMSE)
Connectivity baseline model	3.03	5.35
Local surface model	4.38	12.20
Local volume model	4.41	8.95
Local baseline model	4.13	8.67
	1.94	2.13

The process of the connection feature selection, as well as the model evaluation metrics were included in the Supplementary material. RMSE: Root mean square error; Var: Variance.