Straight to the Heart : Classification of Multi-Channel ECG-signals using MiniROCKET

Abstract: Machine Learning (ML) has revolutionized various domains, with biomedicine standing out as a major beneficiary. In the realm of biomedicine, Convolutional Neural Networks (CNNs) have notably played a pivotal role since their inception, particularly in applications such as time-series classification. Deep Convolutional Neural Networks (DCNNs) have shown promise in classifying electrocardiogram (ECG) signals. However, their deep architecture leads not only to risk for over-fitting when insufficient data is at hand, but also to large computational costs. This study leverages the efficient architecture of Mini-ROCKET, a variant of CNN, to explore improvements in ECG signal classification at Getinge. The primary objective is to enhance the efficiency of the Electrical Activity of the Diaphragm (Edi) catheter position classification compared to the existing Residual Network (ResNet) approach. In the Intensive Care Unit (ICU), patients are often connected to mechanical ventilators operating based on Edi catheter-detected signals. However, weak or absent EMG signals can occur, necessitating ECG interpretation, which lacks the precision required for optimal Edi catheter placement. Clinicians have long recognized the challenges of manual Edi catheter positioning. Currently, positioning relies on manual interpretation of electromyography (EMG) and ECG signals from a 9-lead electrode array. Given the risk for electrode displacement due to patient movements, continuous monitoring by skilled clinicians is essential. This thesis demonstrates the potential of Mini-ROCKET in addressing these challenges. By training the model on Getinge’s proprietary ECG patient dataset, the study aims to measure improvements in computational cost, accuracy, and user value as compared to previous work with Edicathere positioning at Getinge. The findings of this research hold significant implications for the future of ECG signal classification and the broader application of Mini-ROCKET in medical signal processing.