| 摘要: | 本研究在開發一套整合心電圖(Electrocardiogram, ECG)與心音(Phonocardiogram, PCG)的雙模態數位聽診系統，並結合深度學習模型以支援心臟衰竭(Heart Failure)之早期偵測與遠距監測。心臟衰竭為全球高度盛行的慢性疾病，其臨床表現常包含呼吸困難、下肢水腫與活動耐受度下降等，但早期症狀多屬非特異性，臨床評估高度仰賴醫師經驗。ECG 可反映心律、傳導路徑與心肌負荷變化；PCG 則提供瓣膜運動、心室壓力動態與異常血流聲響等機械資訊。若能同步量測 ECG 與 PCG，將有助於更精確地掌握心臟電氣—機械整合狀態，對心衰竭的即時判讀與長期追蹤具有重要價值。
本研究設計一套可攜式、高整合度之硬體平台，包含心音／心電類比前端電路、濾波與偏壓設計、低雜訊電源管理、ADC 數位化架構，以及以 STM32L562 為核心的資料擷取系統。系統採用 SPI 介面搭配雙緩衝技術，成功達成即時資料寫入 SD 卡與 ECG–PCG 時序同步擷取。在心音分析方面，本研究以 YOLOV8 建構心雜音(murmur)偵測模型，並使用 CirCor DigiScope Dataset 進行訓練；同時導入連續與間斷白噪音的資料擴增策略，以增強模型在真實錄音環境下之魯棒性。實驗結果顯示，本系統可穩定取得高品質之 ECG 與 PCG 訊號，並能透過深度學習模型自動標定心雜音區段，於乾淨與高雜訊情境中皆展現良好辨識能力。
綜合以上成果，本研究所提出的雙模態智慧聽診系統兼具便攜性、低功耗與自動化分析能力，可作為心臟衰竭之早期偵測工具，亦適合作為居家監測與遠距醫療的重要基礎平台，具有進一步發展為智慧醫療終端裝置的高度潛力。
;This study presents the development of a dual-modal digital stethoscope that integrates electrocardiogram (ECG) and phonocardiogram (PCG) acquisition, combined with a deep learning–based analysis model to support early detection and remote monitoring of heart failure. Heart failure is a highly prevalent chronic disease worldwide, characterized by symptoms such as dyspnea, peripheral edema, and reduced exercise tolerance. However, its early manifestations are often nonspecific, making clinical assessment highly dependent on physician experience. ECG provides essential information on cardiac rhythm, conduction pathways, and myocardial loading conditions, whereas PCG reflects valvular function, ventricular pressure dynamics, and abnormal blood flow sounds. Simultaneous acquisition of both signals enables a more comprehensive assessment of the heart’s electromechanical behavior, which is crucial for timely interpretation and long-term management of heart failure.
In this work, a portable and highly integrated measurement platform is developed, incorporating analog front-end circuits for PCG and ECG, band-pass filtering and biasing, low-noise power management, analog-to-digital conversion, and an STM32L562-based data acquisition system. The system employs an SPI interface with a double-buffering mechanism to achieve real-time SD-card data logging and reliable time-aligned ECG–PCG acquisition. For heart sound analysis, a YOLOV8-based murmur detection model is constructed using the CirCor DigiScope Dataset, with additional continuous and intermittent white-noise augmentation to improve robustness in realistic acoustic environments. Experimental results demonstrate that the proposed system successfully captures high-quality ECG and PCG signals synchronously and, with the aid of the deep learning model, can automatically identify murmur segments with stable performance under both clean and noisy conditions.
Overall, the proposed dual-modal stethoscope system offers portability, low power consumption, and automated analysis capability. It serves as a promising foundation for early screening of heart failure, home-based physiological monitoring, and remote healthcare applications, with strong potential for future development into an intelligent medical device. |
