Skip to main content

Table 5 Performance of schizophrenic speech detection using classic deep neural networks and the proposed Sch-net

From: Sch-net: a deep learning architecture for automatic detection of schizophrenia

Evaluated indicators 95% CI
AlexNet VGG16 ResNet34 DenseNet121 Xception Sch-net (ours)
Accuracy 0.9272 0.9247 0.9439 0.9469 0.9503 0.9768
(0.9249,0.9295) (0.9225,0.9269) (0.9398,0.9480) (0.9449,0.9489) (0.9482,0.9524) (0.9739,0.9797)
Precision 0.9279 0.8937 0.9074 0.9555 0.9462 0.9639
(0.9226,0.9333) (0.8900,0.8973) (0.9024,0.9124) (0.9516,0.9594) (0.9421,0.9503) (0.9585,0.9693)
Recall 0.9268 0.9643 0.9890 0.9375 0.9551 0.9908
(0.9251,0.9285) (0.9643,0.9643) (0.9822,0.9958) (0.9375,0.9375) (0.9545,0.9556) (0.9898,0.9918)
F1-score 0.9273 0.9276 0.9463 0.9464 0.9506 0.9771
(0.9252,0.9293) (0.9257,0.9295) (0.9423,0.9503) (0.9445,0.9483) (0.9486,0.9526) (0.9743,0.9799)
Sensitivity 0.6399 0.8795 0.9798 0.9262 0.9482 0.9914
(0.6244,0.6554) (0.8715,0.8875) (0.9725,0.9870) (0.9244,0.9280) (0.9409,0.9556) (0.9863,0.9964)
Specificity 0.8747 0.9268 0.9399 0.9938 0.9646 0.9738
(0.8564,0.893) (0.9164,0.9372) (0.9331,0.9467) (0.9910,0.9965) (0.9567,0.9724) (0.9656,0.9820)
AUC 0.7935 0.9447 0.9888 0.9908 0.9924 0.9978
(0.7868,0.8003) (0.9422,0.9472) (0.9855,0.9921) (0.9899,0.9917) (0.9912,0.9936) (0.9965,0.9990)