AdapSQA: Adaptive ECG Signal Quality Assessment Model for Inter-Patient Paradigm using Unsupervised Domain Adaptation

Signa1 quality assessment (SQA) is an important topic in the field of wearable electrocardiogram (ECG) monitoring. Existing ECG SQA models focus on the intra-patient paradigm where the training and testing data are from same individuals. However, due to the individual differences in ECG morphology, features extracted from the the training patients may not be applicable to the new patient. Therefore, these models may suffer severe performance degradation in the inter-patient paradigm which is closer to the reality. In this paper, we propose a novel adaptive ECG SQA model called AdapSQA for the inter-patient paradigm using unsupervised domain adaptation in order to enhance its feature extraction adaptability to the new patient. To realize our AdapSQA, a lightweight baseline model for ECG SQA is first built for better feature extraction in wearable systems. Then, a domain adaptation layer is introduced to align the feature distribution of the training patients and the the new patient by minimizing the distance between the two domains. In this way, a baseline model can be adaptive to a new patient without extra annotation. To evaluate the proposed model, a patient-specific ECG Noise Dataset was generated based on the public datasets since there is no public open source of interest. Experimental results demonstrate that our proposed AdapSQA outperforms state-of-the-art approaches in term of the average inter-patient accuracy to 93.67% with a smaller standard deviation of 4.41%, and is able to achieve lightweight deployment for wearable systems.