心臟衰竭患者的病情變化常伴隨心臟收縮功能與血流動力學狀態改變。傳統每搏輸出量（stroke volume， SV）評估多仰賴心臟超音波或右心導管檢查，雖可提供重要臨床資訊，卻不易應用於長時間居家監測。本研究旨在建立一套整合心電圖（electrocardiogram， ECG）與心震圖（seismocardiography， SCG）的非侵入式心臟功能分析流程，利用嶺迴歸（ridge regression）估測 SV，並實作可量測胸前 SCG 訊號的穿戴式感測裝置。 本研究使用 SCG-RHC 公開資料庫，資料包含胸前穿戴式感測器同步量測的 ECG 與 SCG 訊號，並以右心導管檢查（right heart catheterization， RHC）所得的 SV 作為臨床參考標準。訊號處理流程以 ECG R 波峰進行心搏分段與時間對齊，再由代表性 SCG 心搏偵測主動脈瓣開啟（aortic valve opening， AO）與關閉（aortic valve closure， AC）特徵點。模型輸入包含射血前期指數（pre-ejection period index， PEPI）、心室射血時間（ventricular ejection time， VET）、PEP/VET、VET/PEP、PEP 與 VET 區間的 SCG 能量特徵、穿戴式心率及身體質量指數（body mass index， BMI）。最終共納入 51 筆有效紀錄，並以受試者層級十折交叉驗證評估模型效能。 在 42 組正常 SCG 參考波形中，AO 與 AC 特徵點於 ±20 ms 誤差範圍內的定位準確率分別為 90.5% 與 92.9%，平均絕對誤差分別為 7.4 ms 與 8.2 ms。嶺迴歸模型經受試者層級十折交叉驗證後，均方根誤差為 17.160 mL，Pearson 相關係數為 0.533，決定係數為 0.262。結果顯示，ECG–SCG 特徵與右心導管量測的 SV 呈初步正相關，但模型效能仍須以更大樣本及外部資料進一步驗證。 此外，本研究完成穿戴式心震感測裝置的設計與初步量測。自製裝置取得的胸前 SCG 訊號與外部 ECG 訊號呈現心搏週期上的時間對應關係，並可用於心搏分段、代表性心搏建立及 AO、AC 特徵點分析。綜合而言，本研究建立了結合臨床 ECG–SCG 資料分析、SV 估測與穿戴式 SCG 感測硬體的初步流程，可作為未來心臟衰竭長期追蹤與居家監測研究的基礎。;This study developed a non-invasive cardiac function analysis workflow integrating electrocardiography (ECG) and seismocardiography (SCG) to estimate stroke volume (SV) and implemented a wearable platform for measuring chest SCG signals. Conventional SV assessment mainly relies on echocardiography or right heart catheterization (RHC)， limiting its application in long-term home monitoring. The publicly available SCG-RHC database was used， with invasively measured SV serving as the clinical reference. ECG R peaks were used for heartbeat segmentation and signal alignment， followed by automatic detection of aortic valve opening (AO) and closure (AC) points in the SCG waveform. Cardiac timing and SCG energy features， together with heart rate and body mass index， were used to develop a ridge regression model. A total of 51 valid records were evaluated using subject-level 10-fold cross-validation. Using 42 normal reference SCG waveforms， the AO and AC detection accuracies within ±20 ms were 90.5% and 92.9%， respectively， with mean absolute errors of 7.4 and 8.2 ms. The SV estimation model achieved a root mean square error of 17.160 mL， a Pearson correlation coefficient of 0.533， and a coefficient of determination of 0.262. These findings indicate a preliminary association between ECG–SCG features and invasively measured SV. The wearable platform also successfully acquired periodic chest wall vibrations corresponding to cardiac cycles during approximately synchronized recording with an external ECG device. The acquired signals supported heartbeat segmentation， representative beat construction， and SCG fiducial point analysis. Overall， the proposed workflow provides a foundation for future long-term monitoring and wearable cardiac function assessment in patients with heart failure.
