Figure 3.

The overall processing framework of $\text{NeurDNet}$ to perform differential diagnosis between PD and ET. (a) This part depicts the processing pipeline for the first-stage classifier, which is based on convolutional neural networks. In this stage, a preliminary decision (PD or ET) is made on a single signal of tremor assessment, which is previously passed through the pre-processing block. This signal could be the acceleration of hand motion in any axis, from any task of any trial. (b) This figure shows the second stage of the classification process for each tremor assessment. In fact, each tremor assessment contains 54 tremor signals, where all of them are passed through the first-stage classifier. Then, the decision on each signal is aggregated in a vector of length 54 which forms the feature vector for the second-stage classifier.