Table 5.

External validation.

	Dice	Balanced Accuracy
CNN	0.845 (0.079)	0.874 (0.065)
DNN	0.788 (0.075)	0.836 (0.067)
SVM	0.786 (0.077)	0.832 (0.056)

Mean (SD) Dice index and balanced accuracy of the three machine learning models for DWMA detection. The proposed CNN model outperformed compared DNN and SVM in the external validation with 50 subjects from Siemens scanner as training set and 28 subjects from GE scanner as independent testing set. CNN, convolutional neural networks; DNN, deep neural networks; SVM, support vector machine. The boldface denotes the best performance for individual metrics in the validation experiment.