Since it was first issued in 1964， the Declaration of Helsinki has undergone numerous revisions， aiming to address the ethical challenges in the field of medical research. This paper systematically reviewed the origins and historical evolution of the declaration， as well as sorted out the key revision contents at different stages and their ethical logic in detail. In the face of the challenges of technological revolution and globalization， the 2024 revision provided directional guidance to meet the needs of future medical research while reinforcing ethical norms. Its historical evolution is not only a microcosm of the development of ethics but also reflects the shared responsibility of medical research in the context of globalization.

Algorithms are the operational key for medical artificial intelligence(AI)to optimize the patients seeking medical treatment.With the gradual implementation and application of medical AI algorithms,their decision-making presents the main characteristics of the data-driven,algorithmic black box,and probabilistic paradigms in generating patterns.These characteristics have driven the transformation from artificial diagnosis and treatment to intelligent diagnosis and treatment,but have also simultaneously spawned corresponding practical problems such as"algorithms"replacing doctor decision-making,"algorithmic Leviathan"resolving patients'basic rights,and doctor-patient trust crisis.To effectively prevent ethical risks arising from medical AI algorithm decision-making,it is necessary to call for a return to the subject-centered approach with doctors as the main body and algorithm technology as the wings,strengthen the doctor's decision-making ability,respect patients'basic rights,promote algorithm optimization,train technology to improve,anchor risk keys,and strengthen doctor-patient trust construction.

Objective:To evaluate the accuracy of continuous photoplethysmography algorithms for atrial fibrillation diagnosis and atrial fibrillation burden evaluation via wearable devices.Methods:This study was a self-controlled prospective cohort study. A total of 254 consecutive inpatients were recruited from the Department of Cardiology, Beijing Anzhen Hospital, Capital Medical University from September 2022 to November 2022. All participants were required to wear two devices at the same time: (1) an electrocardiogram (ECG) watch for acquisition of photoplethysmography (watch-recorded-photoplethysmography, W-PPG) and electrocardiogram (watch-recorded-electrocardiogram, W-ECG); (2) an ECG patch for acquisition of electrocardiogram (patch-recorded-electrocardiogram, P-ECG). The results were measured in 30 s data segments and individual participants, separately, which were calculated for sensitivity and specificity by comparing them with the results of expert-read ECG according to the receiver operating characteristic curve. Four groups were separated according to the proportion of the atrial fibrillation burden, and the difference of atrial fibrillation burden from algorithm and expert-read ECG was calculated.Results:All 254 subjects completed the study. The mean age of participants was (63.04±11.04) years old, 99 (38.98%) of them were female, and 97 (38.19%) patients had persistent atrial fibrillation. Expert-read ECG results were taken as standard criteria in all calculations. The P-ECG algorithm had a sensitivity of 94.86% (95% CI: 94.81%-94.91%) and a specificity of 99.30% (95% CI: 99.28%-99.31%) when measured in data segments. The W-PPG algorithm had a sensitivity of 96.07% (95% CI: 95.97%-96.18%) and a specificity of 98.62% (95% CI: 98.59%-98.65%). When measured in terms of individual participants, the P-ECG algorithm had a sensitivity of 92.55% (95% CI: 87.57%-95.71%) and a specificity of 96.39% (95% CI: 93.45%-98.09%), while the W-PPG algorithm had a sensitivity of 93.71% (95% CI: 88.75%-96.67%) and a specificity of 89.62% (95% CI: 85.61%-92.65%). When measured in terms of a single acquisition of W-ECG records, the W-ECG algorithm had a sensitivity of 92.04% (95% CI: 88.14%-94.78%) and a specificity of 96.19% (95% CI: 94.35%-97.47%). For atrial fibrillation burden assessment, the difference between the W-PPG analysis results and the expert-read ECG results was less than 2% in all burden distribution intervals. Conclusions:Continuous photoplethysmography algorithm applied to wearable devices to detect atrial fibrillation is a feasible strategy. Taking expert-read ECG results as standard, continuous monitoring of PPG by a smartwatch is highly accurate for atrial fibrillation diagnosis and can also be used to effectively complete atrial fibrillation burden assessment.