Estimating preoperative blood volume from ultrasound using deep learning

Quantitative Estimation of Preoperative Blood Volume Using Multi-modal Ultrasound and Deep Learning

This project will test whether a deep learning model can use ultrasound videos to estimate blood volume in adults before surgery.

Quick facts

• Study type: Observational
• Enrollment: 800 (estimated)
• Ages: 18 Years to 75 Years
• Sex: All
• Sponsor: Shanghai 6th People's Hospital (other)
• Locations: 2 sites (Shanghai, Shanghai Municipality and 1 other locations)
• Trial ID: NCT06957587 on ClinicalTrials.gov

What this trial studies

This is a prospective, single-center observational study enrolling adult patients scheduled for surgery to collect standardized preoperative ultrasound video clips of major vessels. Researchers will combine multi-modal dynamic ultrasound of the internal jugular, subclavian, inferior vena cava, and common carotid with relevant clinical data to train and validate a deep learning model to estimate blood volume. Eligible participants are adults 18–75 years old with BMI 18–30 kg/m2 and ASA physical status I–II, while patients with significant organ dysfunction, severe anemia, poor ultrasound windows, multiple trauma, or pregnancy are excluded. Model performance will be compared against available clinical reference measures to quantify accuracy and reliability in a real-world preoperative population.

Who should consider this trial

Good fit: Adults aged 18–75 with BMI 18–30 kg/m2, ASA I–II, who are scheduled for surgery and can undergo standard ultrasound imaging are ideal candidates.

Not a fit: Patients with poor ultrasound visualization of target vessels, significant cardiac/respiratory/liver/kidney dysfunction, hemoglobin <10 g/dL, multiple major injuries, or who are pregnant are unlikely to benefit from the method tested here.

Why it matters

Potential benefit: If successful, this approach could provide a quick, non-invasive way to quantify blood volume before surgery and help guide fluid and transfusion decisions.

How similar studies have performed: Machine-learning methods have shown promise for extracting information from ultrasound, but direct non-invasive quantitative blood volume estimation from multi-vessel ultrasound remains largely novel and unproven.

Eligibility criteria

Inclusion Criteria:

* Agree to join this study and sign the informed consent form;
* Age between 18 and 75 years old (inclusive);
* BMI (body mass index) is between 18 and 30 kg/m2;
* American Society of Anesthesiologists (ASA) grades I-II

Exclusion Criteria:

* Preoperative hemoglobin (Hb) \<10g/dl
* Cardiac dysfunction (NYHA class III-IV), respiratory dysfunction (ATS class 2-4), history of liver and kidney dysfunction (such as transaminase / albumin / bilirubin abnormalities, hepatitis history, serum creatinine / urea nitrogen rise, etc.), nervous system abnormalities (those who cannot cooperate due to stroke or its sequelae, Alzheimer, etc.);
* The ultrasonic display of inferior vena cava, internal jugular vein, subclavian vein or common carotid artery is extremely poor, venous thrombosis or anatomical abnormalities;
* Multiple injury with chest, abdomen or brain;
* Pregnant woman

Where this trial is running

Shanghai, Shanghai Municipality and 1 other locations

• Shanghai Jiao Tong University Affiliated Sixth People's Hospital — Shanghai, Shanghai Municipality, China (NOT_YET_RECRUITING)
• Shanghai Jiao Tong University Affiliated Sixth People's Hospital — Shanghai, Shanghai Municipality, China (RECRUITING)

Study contacts

• Study coordinator: xiuxiu sun, MD

Conditions: Blood Volume Analysis, Ultrasound, Machine Learning