Prediction of amyloid PET positivity via machine learning algorithms trained with EDTA-based blood amyloid-β oligomerization data

Table 2 The mean performance of the MDS-OAβ predicting amyloid PET positivity, evaluated using various machine learning algorithms on 50 trials (mean ± standard deviation %)

Algorithms	Performance	MDS-OAβ	MDS-OAβ + Age	MDS-OAβ + Age + APOE	MDS-OAβ + Age + MMSE	MDS-OAβ + Age + MMSE + APOE
Subject number		N = 312	N = 312	N = 263	N = 289	N = 246
Support vector machine	Acc	71.09 ± 3.27**	69.21 ± 4.07	68.76 ± 3.99	68.69 ± 4.02	69.86 ± 4.82
	Prec	80.06 ± 4.46	76.70 ± 4.03	76.72 ± 4.50	78.25 ± 3.71	82.22 ± 5.25
	Rec	80.76 ± 5.38	83.13 ± 5.54	82.99 ± 4.80	78.93 ± 6.69	76.84 ± 5.68
	F1-value	80.18 ± 2.70	79.61 ± 3.05	79.59 ± 3.04	78.36 ± 3.45	79.24 ± 3.82
Random forest	Acc	66.08 ± 4.15	67.75 ± 3.61	69.49 ± 4.01	75.54 ± 3.98*	77.14 ± 4.21*†
	Prec	77.28 ± 4.61	75.68 ± 4.93	76.72 ± 5.54	79.84 ± 4.56	80.75 ± 4.65
	Rec	75.93 ± 5.57	82.17 ± 5.17	84.62 ± 4.56	89.81 ± 3.76	91.05 ± 4.78
	F1-value	76.40 ± 3.27	78.59 ± 3.08	80.26 ± 2.95	84.42 ± 2.92	85.44 ± 3.10
Logistic regression	Acc	69.13 ± 3.91**	69.00 ± 4.06	69.19 ± 4.98	69.38 ± 4.72	73.96 ± 5.30
	Prec	71.56 ± 3.78	73.33 ± 4.48	74.15 ± 5.27	75.59 ± 5.85	80.84 ± 5.21
	Rec	94.22 ± 3.76	90.30 ± 4.85	89.31 ± 7.41	86.56 ± 6.10	85.58 ± 5.99
	F1-value	81.25 ± 2.81	80.77 ± 2.93	80.72 ± 3.81	80.38 ± 3.35	82.94 ± 3.79
Deep neural network	Acc	64.00 ± 4.50	64.83 ± 4.45	64.50 ± 4.77	66.80 ± 5.16	69.24 ± 4.18†
	Prec	80.81 ± 4.90	77.19 ± 4.37	76.60 ± 4.82	78.50 ± 4.41	80.52 ± 4.12
	Rec	66.39 ± 8.25	74.20 ± 6.25	75.06 ± 6.31	75.65 ± 7.00	78.03 ± 6.18
	F1-value	72.46 ± 4.87	75.45 ± 3.64	75.61 ± 3.94	76.81 ± 4.04	79.03 ± 3.19

MDS-OAβ, Multimer Detection System-Oligomeric Amyloid-β; APOE, apolipoprotein E; Acc, accuracy; Prec, precision; Rec, recall
*p = 0.054, when compared ‘MDS-OAβ + Age + MMSE’ with ‘MDS-OAβ + Age + MMSE + APOE’
**p < 0.01, when compared ‘MDS-OAβ’ only of Support Vector Machine model with Logistic Regression
^†p < 0.001, when compared ‘Random Forest’ with ‘Deep Neural Network’ algorithm based on the Student t-test

ISSN: 1472-6947