Table 6.

Comparison of average classification performance for multiscale convolutional mixture of experts and wavelet-based convolutional mixture of experts on the research Heidelberg datasets based on 10 repetitions of 5-fold cross-validation method with τ=15% for decision-making

Model	Configuration	Dataset	Curvature correction	Average λ	Performance
					Precision (%)	AUC	Average testing time (s/VOI)
MCME[16]	l3−l2−l1−l0	Set 1	×	0.08	94.68±1.47	0.961	10.9
			√	0.21	99.01±0.41	0.998
		Set 2	×	0.17	94.63±2.45	0.974
			√	0.26	97.11±0.96	0.994
WCME[17]	LL−LH−HL−HH	Set 1	×	0.12	95.91±1.56	0.976	9.34
			√	0.18	97.40±0.78	0.995
		Set 2	×	0.21	96.03±2.14	0.986
			√	0.24	96.33±1.29	0.990

MCME – Multiscale convolutional mixture of experts; AUC – Area under the ROC curve; ROC – Receiver operating characteristic; WCME – Wavelet-based convolutional mixture of experts; VOI – Volume of interest