Toward Embedded Multi-Level Classification of 12-Lead ECG Signal Quality Using Spectrograms and CNNs

This study presents an open and replicable methodology for multi-lead ECG signal quality assessment (SQA), implemented on a 12-lead embedded acquisition platform. Signal quality is a critical software component for diagnostic reliability and compliance with international standards such as IEC 60601-2-27 (clinical ECG monitors), IEC 60601-2-47 (ambulatory ECG systems), and IEC 62304 (software life cycle for medical devices) which define the essential engineering requirements and functional performance for medical devices. Unlike proprietary SQA algorithms embedded in closed commercial systems such as Philips DXL™, the proposed method provides a transparent and auditable framework that enables independent validation and supports adaptation for research and clinical prototyping. Our approach combines convolutional neural networks (CNNs) with FFT-derived spectrograms to perform four-level signal quality classification (High, Medium, Low, and Unidentifiable), achieving up to 95.67% accuracy on the test set, confirming the robustness of the CNN-based spectrogram classification model. The algorithm has been validated on a custom controlled dataset generated using the Fluke PS420™ hardware simulator, enabling controlled replication of signal artifacts for software-level evaluation. Designed for execution on resource-constrained embedded platforms, the system integrates real-time preprocessing and wireless transmission, demonstrating its feasibility for deployment in mobile or decentralized ECG monitoring solutions. These results establish a software validation proof-of-concept that goes beyond algorithmic performance, addressing regulatory expectations such as those outlined in FDA’s Good Machine Learning Practice (GMLP). While clinical validation remains pending, this work contributes a standards-aligned methodology to democratize advanced SQA functionality and support future regulatory-compliant development of embedded ECG system.