An Automated Machine Learning Method to Efficiently Classify the 12-lead ECG Signal Acquisition Quality

Given their low cost and non-invasive nature, ElectroCardioGram (ECG) signals have been widely used as a useful tool for diagnosing heart diseases. However, acquisition issues such as electrode interchange and oscillation noise may negatively impact expert exam interpretation and even automatic classification tasks. Here we propose an automated machine learning method to efficiently classify the 12-lead ECG signal acquisition quality. It consists of a two-stage classification process. Firstly, the ECG signals are processed and segmented aiming to classify them as noisy or acceptable signals. Then, the second classification stage yields the binary classification correct acquisition or limb electrodes interchange. Concerning the electrode positioning, the Random Forest technique presented interesting results (precision of 97%, recall of 89%, and F1-Score of 93%). Concerning noise detection, Random Forest presented a general accuracy of 85%, a recall of 57%, and a precision of 91%. All the obtained results yield to consider the proposed framework for application within a real telemedicine environment.