Development of an AI-based Emergency Imaging Multi-Disease Rapid Joint Screening System

Not yet recruiting

• Conditions: Emergency Service, Hospital; Diagnosis; Emergency Medical Services; Critical Illness; Machine Learning
• Interventions: Diagnostic Test: radiomic of CT

• Registration Number: NCT05974163
• Lead Sponsor: Sun Yat-Sen Memorial Hospital of Sun Yat-Sen University

Brief Summary

Introduction:

Early and rapid diagnosis of etiology is often an important part of saving the lives of patients in emergency department. Chest CT is an important examination method for emergency diagnosis because of its fast examination speed and accurate localization. Traditional medical imaging diagnosis relies on radiologists to report in a qualitative and subjective manner. Through the interdisciplinary combination of clinical, imaging and artificial intelligence, the integration of multi-omics data, the construction of large-scale language models, and the construction of the auxiliary diagnosis support system of "one check for multiple diseases" provide new ideas and means for the rapid and accurate screening of emergency critical diseases.

Method：

Study design Investigators retrospectively collected cardiovascular, respiratory, digestive, and neurological CT images, demographic data, medical history and laboratory date of emergency department patients during the period from 1 January 2018 and 30 December 2024. Regularly carry out standardized follow-up work, and complete the collection and database establishment of clinical-imaging multi-omics data of patients attending emergency department.The inclusion criteria are:1. adult emergency patients with cardiovascular, respiratory, digestive, and nervous system diseases; 2. These patients had CT images. Patients with incomplete clinical or radiographic data were excluded from the analysis. Regularly carry out standardized follow-up work, and complete the collection and database establishment of clinical-imaging multi-omics data of patients attending emergency department.

Based on the collected medical text data, an artificial intelligence large-scale language model algorithm framework is built. After the structure annotation of chest CT images is performed by doctors above the intermediate level of imaging, the Transformer deep neural network is trained for CT image segmentation, and a series of tasks such as structural structure segmentation, damage detection, disease classification and automatic report generation are developed based on Vision Transformer self-attention architecture mechanism. A multi-disease diagnosis and treatment decision-making system based on chest CT images, clinical text and examination multimodal data was constructed and validated.

Disscusion

Emergency medicine deals mainly with unpredictable critical and sudden illnesses. Patients who come to the emergency department for medical treatment often have acute onset, hidden condition, rapid progress, many complications, high mortality and disability rate. Assisted diagnosis systems developed by combining clinical text, images and artificial intelligence can greatly improve the ability of emergency department doctors to accurately diagnose diseases. This study fills the blank of CT artificial intelligence aided diagnosis system for emergency patients, and provides a rapid diagnosis scheme for multi-system and multi-disease. Finally, the results will be transformed into clinical application software and used and promoted in clinical work to improve the diagnosis and treatment level.

Detailed Description

Not available

Study Timeline

• Status Verified: 2023-07
• Study First Submitted: 2023-07-26
• Study First Submitted QC: 2023-07-26
• Last Update Submitted: 2023-07-26
• Study Start: 2023-08-01
• Study First Posted: 2023-08-03
• Last Update Posted: 2023-08-03
• Primary Completion: 2024-07-31
• Study Completion: 2025-07-31

Recruitment & Eligibility

• Status: NOT_YET_RECRUITING
• Sex: All
• Target Recruitment: 10000

Inclusion Criteria

Adults with cardiovascular, respiratory, digestive, and neurological disorders. CT imaging was available.

Exclusion Criteria

Patients with incomplete clinical or radiographic data were excluded.

Study & Design

• Study Type: OBSERVATIONAL
• Study Design: Not specified

Arm && Interventions

Group	Intervention	Description
External validation cohort 2	radiomic of CT	1000 patients will be recruited prospectively during the period from January 2023 to December 2025 as external validation group
Model reconstruction cohort	radiomic of CT	8000 patients were recruited retrospectively from January 2023 to December 2025 as discovering group.
External Validation cohort 1	radiomic of CT	1000 patients were recruited retrospectively from January 2023 to December 2025 as internal validation group.

Primary Outcome Measures

Name	Time	Method
Accuracy of disease diagnosis	2025-08-01~2025-12-31	Construct a rapid diagnosis, accurate and efficient emergency CT image multi-disease rapid joint screening system

Trial Locations

Locations (1)

Sun Yat-sen Memorial Hospital, Sun Yat-sen University; 🇨🇳 Guangzhou, China