This is a prospective, multicenter study designed to validate a deep learning model for screening valvular heart diseases using routine, non-contrast chest computed tomography (CT) scans.
The primary objective is to evaluate the model's diagnostic performance, with the sensitivity serving as the primary efficacy endpoint. Secondary endpoints will include other performance metrics such as area under the receiver operating characteristic curve (AUC), specificity, and accuracy, etc.
Age ≥ 18 years.
Complete electronic health record.
Non-contrast chest CT performed between Nov 1, 2025 - Nov 1, 2026 in any medical context (including physical exam, outpatient, inpatient, or emergency).
AI-predicted moderate or severe valvular heart disease, or deemed to require clinical intervention, or selected negative cases from sampling verification.
Poor-quality non-contrast chest CT images.
Incomplete clinical records, involving severe deficiencies in critical diagnostic results, treatment records, imaging data, surgical records, medical history summaries, laboratory test results, or other essential medical information.
Presence of prosthetic valve implants, including aortic valves (mechanical valves, bioprosthetic valves), mitral valves (transcatheter edge-to-edge repair, bioprosthetic valves, mechanical valves, annuloplasty rings), tricuspid valves (TEER clipping, bioprosthetic valves, mechanical valves, annuloplasty rings), pulmonary valves (bioprosthetic valves), etc.
Abnormalities or conditions deemed by the investigator to warrant exclusion from the study enrollment.