Table 2.

Comparison of classification performance over different experiments with multi-type features.

	Accuracy	NC sen.	AD sen.	FTD sen.
PCA+SVM (Multitype)	83.06	93.90	71.74	68.23
MDNN+Thickness	83.04	89.07	76.77	74.79
MDNN+Volume	85.97	91.05	83.88	74.20
MMDNN (Multitype)	86.81	93.76	81.94	73.59
GAN (Multitype)	88.28	93.40	84.66	77.82

The second column is the overall classification accuracy, while the third to fifth columns represent the sensitivity of NC, AD, and FTD, respectively. The second row represents the result with PCA+SVM using multi-type multi-scale features. The third and fourth rows are the classification performance of MDNN with multi-scale surface thickness or multi-scale volume size. Experiment results with both types of features using MMDNN are shown in the fifth row and the last row represents the result of multi-type multi-scale features along with data augmentation using GAN.