Abstract

CONTINUOUS CARDIAC MONITORING HAS BEEN INCREASINGLY ADOPTED TO PREVENT HEART DISEASES, ESPECIALLY THE CASE OF CHAGAS DISEASE, A CHRONIC CONDITION THAT CAN DEGRADE THE HEART CONDITION, LEADING TO SUDDEN CARDIAC DEATH. UNFORTUNATELY, A COMMON CHALLENGE FOR THESE SYSTEMS IS THE LOW-QUALITY AND HIGH LEVEL OF NOISE IN ECG SIGNAL COLLECTION. ALSO, GENERIC TECHNIQUES TO ASSESS THE ECG QUALITY CAN DISCARD USEFUL INFORMATION IN THESE SO-CALLED CHAGASIC ECG SIGNALS. TO MITIGATE THIS ISSUE, THIS WORK PROPOSES A 1D CNN NETWORK TO ASSESS THE QUALITY OF THE ECG SIGNAL FOR CHAGASIC PATIENTS AND COMPARE IT TO THE STATE OF ART TECHNIQUES. SEGMENTS OF 10 S WERE EXTRACTED FROM 200 1-LEAD ECG HOLTER SIGNALS. DIFFERENT FEATURE EXTRACTIONS WERE CONSIDERED SUCH AS MORPHOLOGICAL FIDUCIAL POINTS, INTERVAL DURATION, AND STATISTICAL FEATURES, AIMING TO CLASSIFY 400 SEGMENTS INTO FOUR SIGNAL QUALITY TYPES: ACCEPTABLE ECG, NON-ECG, WANDERING BASELINE (WB), AND AC INTERFERENCE (ACI) SEGMENTS. THE PROPOSED CNN ARCHITECTURE ACHIEVES A $$0.90 PM 0.02$$ACCURACY IN THE MULTI-CLASSIFICATION EXPERIMENT AND ALSO $$0.94 PM 0.01$$WHEN CONSIDERING ONLY ACCEPTABLE ECG AGAINST THE OTHER THREE CLASSES. ALSO, WE PRESENTED A COMPLEMENTARY EXPERIMENT SHOWING THAT, AFTER REMOVING NOISY SEGMENTS, WE IMPROVED MORPHOLOGICAL RECOGNITION (BASED ON QRS WAVE) BY 33% OF THE ENTIRE ECG DATA. THE PROPOSED NOISE DETECTOR MAY BE APPLIED AS A USEFUL TOOL FOR PRE-PROCESSING CHAGASIC ECG SIGNALS.