. Author manuscript; available in PMC: 2020 Oct 1.

Published in final edited form as: Predict Intell Med. 2019 Oct 10;11843:1–10. doi: 10.1007/978-3-030-32281-6_1

Table 2:

Summary of benchmarks and top-3 methods used in the TADPOLE submissions. DPM – disease progression model.

Submission	Extra^† Features	Nr. of features	Missing data imputation	Diagnosis prediction	ADAS/Vent. prediction
Frog	most features	70+420^*	none	gradient boosting	gradient boosting
EMC1-Std	MRI, ASL, cognitive	250	nearest neighbour	DPM SVM 2D-spline	DPM 2D-spline
VikingAI-Sigmoid	MRI, cognitive, tau	10	none	DPM + ordered logit	DPM
BenchmarkLastVisit	-	3	none	constant model	constant model
BenchmarkME-APOE	APOE	4	none	Gaussian model	linear mixed effects model
BenchmarkSVM	age, APOE	6	mean of previous values	SVM	support vector regressor

^(†)

Aside from the three target biomarkers

^(*)

Augmented features: e.g. min/max, trends, moments.