BrainHemoAI: a deep-learning decision model to guide care for hemorrhagic stroke

Construction of an Integrated Intelligent Model for Spontaneous Intracerebral Hemorrhage Based on Deep Learning

This project will test whether the BrainHemoAI deep-learning model can help doctors use non-contrast CT scans to guide care for people aged 8 and older with spontaneous hemorrhagic stroke.

Quick facts

• Study type: Observational
• Enrollment: 7100 (estimated)
• Ages: 8 Years and up
• Sex: All
• Sponsor: Second Affiliated Hospital of Nanchang University (other)
• Locations: 1 site (Nanchang)
• Trial ID: NCT07570680 on ClinicalTrials.gov

What this trial studies

This observational project will develop and apply a deep-learning decision model using clinical data and non-contrast CT images from patients with spontaneous hemorrhagic stroke treated at participating hospitals. The approach aims to move beyond the traditional Tada formula by combining imaging features and routine clinical variables to reduce subjective differences between clinicians. Data are collected retrospectively and prospectively under standard clinical care, excluding patients who had prior surgery elsewhere or who are not expected to complete follow-up. Collaborating centers include regional hospitals and a university partner to build a larger, more consistent dataset for model training and validation.

Who should consider this trial

Good fit: Ideal candidates are people aged 8 or older with spontaneous hemorrhagic stroke who received non-contrast CT in the outpatient or emergency setting, were treated according to standard clinical guidelines, and have complete clinical data available.

Not a fit: Patients who had prior surgical treatment at another hospital, were in shock on admission, had severe life‑threatening organ dysfunction, died during hospitalization, or cannot complete follow-up are unlikely to benefit from this project.

Why it matters

Potential benefit: If successful, the model could improve the accuracy and consistency of preoperative CT-based decision-making, potentially reducing death and disability and improving recovery after hemorrhagic stroke.

How similar studies have performed: Previous AI work has shown promise for detecting and measuring intracranial hemorrhage on CT, but using deep-learning decision models to standardize surgical or preoperative decision-making remains relatively novel and not yet widely proven.

Eligibility criteria

Inclusion Criteria:

1. Age \>= 8 years old;
2. Patients diagnosed with spontaneous hemorrhagic stroke based on medical history and auxiliary examinations;
3. Received non-contrast computed tomography (NCCT) in the outpatient or emergency department;
4. Treated in accordance with standard clinical guidelines during hospitalization;
5. Have complete clinical data.

Exclusion Criteria:

1. Had undergone surgical treatment in another hospital before admission;
2. Was in a state of shock upon admission;
3. Had severe heart, liver, or kidney dysfunction or other life-threatening systemic diseases;
4. Died during hospitalization;
5. Had an expected lifespan of less than six months or was unable to complete the study follow-up for other reasons.

Where this trial is running

Nanchang

• The Second Affiliated Hospital of Nanchang University — Nanchang, China (RECRUITING)

Study contacts

• Study coordinator: Xingen Zhu, Prof

Conditions: Hemorrhage Stroke