AI-assisted handheld echocardiography for hospital bedside cardiac triage: the prospective OPTIMUST implementation study

A governed AI-enabled handheld ultrasound pathway was feasible in a tertiary hospital, with 71.8% of 287 attempts producing analyzable examinations when performed by non-certified physicians after structured training. Prospective implementation data integrating AI, structured training, digital workflows, and expert governance were previously lacking; this study supplies evidence for that integrated in-hospital pathway. Prospective single-center implementation study with 287 attempted examinations, feasibility as a co-primary endpoint measured by analyzable examinations, and reported rates of non-analyzable (17.4%) and incomplete-report (10.8%) studies.

Operator assessment of the same handheld image sets correlated with centralized expert review for LVEF (r = 0.84), and filling-pressure category agreement yielded a quadratic weighted κ of 0.659. Prospective quantitative data on operator-versus-expert interpretive agreement for the same handheld image sets in real ward settings were previously limited. Agreement analyses were based on analyzable studies; LVEF was reported with a correlation coefficient and Bland-Altman limits of agreement of approximately -21 to +19 percentage points, filling pressure with a quadratic weighted κ, and the comparison was against expert review of the same handheld images rather than comprehensive echocardiography.

Digital integration enabled centralized archiving, structured reporting, remote review, and quality assurance, and physicians reported that handheld findings changed or confirmed management in 93.7% of examinations. Connecting handheld ultrasound to institutional archiving and electronic medical records with remote review and quality assurance is a workflow element less systematically described in prior implementation reports. Workflow integration is described descriptively; the 93.7% clinical impact figure is physician-reported and not independently adjudicated, and 32.8% underwent comprehensive echocardiography within one month.

The study addresses bedside cardiac triage on cardiology and non-cardiology wards of a tertiary hospital, applies to physicians who completed structured training without formal echocardiography certification, and depends on digital governance conditions such as institutional archiving, structured reporting, remote review, and quality assurance; its conclusions are positioned around implementation feasibility and operator-expert agreement on the same handheld image sets rather than replacement of comprehensive echocardiography.

The study does not establish safety, diagnostic accuracy against comprehensive echocardiography, or an independent effect of AI; the 93.7% clinical impact figure is physician-reported and not independently adjudicated, and 32.8% underwent comprehensive echocardiography within one month; moreover, this reading is based on summary-level text, so the specific figures and tables behind image-quality grading and agreement analyses were not included, and assessing the robustness of the image-quality distribution and agreement limits would still require the original figures and tables.