| Title |
Diagnosis-specific Multi-model Design for 12-lead ECG Multi-label Classification |
| Authors |
차재빈(Jae-Bin Cha) ; 황서림(Seo-Rim Hwang) ; 박영철(Young-Cheol Park) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.8.39 |
| Keywords |
ECG(Electrocardiogram); Deep learning; Multi-label classification; Multi-model |
| Abstract |
This paper designed a multi-model that performs multi-label classification by diagnosis using 12-lead electrocardiogram (ECG) signals. The proposed multi-model classifies each ECG signal according to rhythm, duration, amplitude, and morphology, and for this purpose, diagnoses that share similar classification criteria are grouped. At this time, the Minnesota Code Manual, which is widely used as an ECG reading equipment interpretation algorithm, was referred to for group diagnoses, and a large-scale open source database of 45,152 records was used for the experiment. We used various objective metrics such as AUPRC, F1-score, Precision, Recall, and Specificity to evaluate the experimental results. As a result, the proposed multi-model showed excellent performance in all evaluation metrics compared to the baseline single model. In particular, the proposed multi-model showed an improved performance of 0.038 in AUPRC and 0.036 in F1-score on average compared to the baseline single model. |