Machine learning, medical diagnosis, and biomedical engineering research - commentary

Table 1 Impact of the operator and the device type on the sensitivity and specificity of the classifier SVM

In this case, the SVM had been trained on 100 images (50 healthy, 50 ill patients) acquired by Operator 1 and validated using independent sets of the same size of images obtained by Operator 1 or 2. The performance of the classifier is generally better with Operator 1 on data collected on device 1 (which originated the data) although there are variations in performance. Both operators used both devices to acquire data.