Development and Validation of a Deep Learning–Based Automatic Brain Segmentation and Classification Algorithm for Alzheimer Disease Using 3D T1-Weighted Volumetric Images

C.H. Suh; W.H. Shim; S.J. Kim; J.H. Roh; J.-H. Lee; M.-J. Kim; S. Park; W. Jung; J. Sung; G.-H. Jahng,; for the Alzheimer’s Disease Neuroimaging Initiative

doi:10.3174/ajnr.A6848

Article Figures & Data

Figures

Tables

Download figure
Open in new tab
Download powerpoint
FIG 1.
Network architecture of the brain parcellation and classification model. CONV indicates convolution.
Download figure
Open in new tab
Download powerpoint
FIG 2.
The impact of feature (volume of each brain region) on the AD prediction model, as represented by Shapley values, in which the impact of a feature is defined as the change in the expected output of the model when a feature is observed versus unknown. A, Visualization of the top 5 brain regions representing feature impacts pushing the decision of the model to AD, along with average feature impact. B, Visualization of the top 5 brain regions representing feature impacts pushing the decision of the model to healthy controls, along with average feature impact. Bankssts indicates banks of the superior temporal sulcus; SHAP, Shapley Additive Explanations (https://pbiecek.github.io/ema/shapley.html).

Tables

Figures

View popup

Table 1:

Characteristics of the development and datasets^{^a}

	Asan Medical Center	Kyung Hee University Hospital in Gangdong	ADNI	OASIS
No. of patients	1099	212	711	705
Age (mean) (yr)	65 ± 13	70 ± 9	76 ± 7	68 ± 10
No. of male patients	500 (45)	52 (25)	412 (58)	403 (57)
No. of female patients	599 (55)	160 (75)	299 (42)	302 (43)
Classification
AD	161 (15)	68 (32)	178 (25)	145 (21)
Education (yr)	9.9 (4.8)	NA	14.5 (3.4)	14.0 (3.2)
MMSE score	18.5 (4.7)	17.4 (5.3)	22.8 (3.1)	24.4 (5.1)
Clinical Dementia Rating	1.00 (0.49)	1.10 (0.47)	0.73 (0.34)	0.68 (0.28)
Global Deterioration Scale	NA	NA	1.7 (1.4)	3.2 (7.3)
MCI	363 (33)	63 (30)	317 (45)	0
Education (yr)	10.1 (5.0)	NA	15.9 (2.5)
MMSE score	24.9 (3.6)	25.7 (3.7)	26.4 (2.1)
Clinical Dementia Rating	0.51 (0.09)	0.61 (1.16)	0.5
Global Deterioration Scale	NA	NA	1.5 (1.3)
Healthy control	575 (52)	81 (38)	216 (30)	560 (79)
Education (yr)	NA	NA	16.2 (2.8)	15.2 (2.7)
MMSE score	29.5 (0.5)	27.7 (2.5)	29.1 (1.0)	28.8 (3.2)
Clinical Dementia Rating	NA	0.24 (0.26)	0	0
Global Deterioration Scale	NA	NA	0.8 (1.1)	1.3 (4.0)

Note:—MMSE indicates Mini-Mental State Examination; NA, not available.
↵a Unless otherwise indicated, data are reported as number (%).

View popup

Table 2:

Diagnostic performance of logistic regression, the linear Support Vector Machine, and the deep learning–based automatic classification algorithm in the datasets^{^a}

	Logistic Regression	Linear SVM	XGBoost	P Value^{^b}	P Value^{^c}
AD vs MCI
Asan Medical Center	0.770 (0.761–0.779)	0.772 (0.761–0.782)	0.803 (0.802–0.805)	<.001	<.001
Kyung Hee University Hospital at Gangdong	0.798 (0.775–0.822)	0.804 (0.783–0.824)	0.825 (0.810–0.840)	.018	.030
ADNI	0.706 (0.702–0.710)	0.700 (0.695–0.704)	0.758 (0.755–0.760)	<.001	<.001
MCI vs healthy control
Asan Medical Center	0.812 (0.806–0.817)	0.830 (0.821–0.840)	0.870 (0.868–0.872)	<.001	<.001
Kyung Hee University Hospital at Gangdong	0.692 (0.678–0.706)	0.687 (0.669–0.706)	0.705 (0.699–0.712)	.029	.023
ADNI	0.698 (0.686–0.710)	0.702 (0.697–0.708)	0.668 (0.664–0.671)	<.001	<.001
AD vs healthy controls
Asan Medical Center	0.953 (0.949–0.958)	0.960 (0.958–0.963)	0.982 (0.980–0.985)	<.001	<.001
Kyung Hee University Hospital at Gangdong	0.905 (0.889–0.921)	0.911 (0.903–0.920)	0.940 (0.933–0.947)	<.001	<.001
ADNI	0.863 (0.856–0.870)	0.860 (0.857–0.863)	0.885 (0.879–0.891)	<.001	<.001
OASIS^{^d}	0.826 (0.817–0.835)	0.820 (0.809–0.832)	0.840 (0.837–0.844)	.001	<.001

↵a Data are AUC (95% CI).
↵b P values: between logistic regression and XGBoost.
↵c P values: between linear SVM and XGBosst.
↵d OASIS dataset included only AD and healthy controls.