Validation of a Multimodal Artificial Intelligence Model in in Diagnosing Pancreatic Solid Lesions: a Prospective, Multicenter, Randomized, Controlled Trial
Trial Snapshot
- Phase
- Not Applicable
- Status
- Not yet recruiting
- Enrollment
- 716
- Locations
- 1
- Primary Endpoint
- Rate of correct diagnostic classification with assistance of the Joint-AI Model
Study Overview
Brief Summary
This clinical trial aims to learn if a multimodal artificial intelligence (AI) model can enhance the diagnosis of pancreatic solid lesions. The main questions it aims to answer are:
- Does the AI model enhance the diagnostic performance of endoscopists in diagnosing pancreatic solid lesions?
- Does the addition of interpretability analysis further improve the diagnostic performance of the assisted endoscopists? Researchers will compare the diagnostic performance of endoscopists with or without the assistance of the AI model.
Participants will:
- Their clinical data will be prospectively collected.
- They will be randomized to the AI-assist group and the conventional diagnosis group.
Detailed Description
The investigators have previously developed a multimodal AI model (Joint-AI) based on endoscopic ultrasound images and clinical data to diagnose pancreatic solid lesions. This study aims to improve the Joint-AI model's performance with a prospectively collected dataset and validate it through a randomized controlled clinical trial.
Study Design
- Study Type
- Interventional
- Allocation
- Randomized
- Intervention Model
- Parallel
- Primary Purpose
- Diagnostic
- Masking
- Double (Participant, Outcomes Assessor)
Masking Description
During the endoscopic ultrasound procedure, the allocation of participants will be masked to the endoscopists
Eligibility Criteria
- Ages
- 18 Years to — (Adult, Older Adult)
- Sex
- All
- Accepts Healthy Volunteers
- No
Inclusion Criteria
- •Imaging examinations (MRI, CT, B-ultrasound) show a solid mass in the pancreas, which requires endoscopic ultrasound guided-fine needle aspiration/biopsy (EUS-FNA/B) to clarify the nature of the lesion in patients.
- •Written consent provided
Exclusion Criteria
- •Age under 18 years old
Arms & Interventions
Conventional diagnosis
Endoscopists diagnose pancreatic solid lesions according to endoscopic ultrasound images and clinical data.
Joint-AI assisted diagnosis
Endoscopists diagnose pancreatic solid lesions based on endoscopic ultrasound images, clinical data, and predictions made by the Joint-AI model.
Intervention: The assistance of the Joint-AI model (Diagnostic Test)
Interpretable Joint-AI assisted diagnosis
Endoscopists diagnose pancreatic solid lesions based on endoscopic ultrasound images, clinical data, predictions given by the Joint-AI, and interpretability analysis results used to improve the transparency of the decision-making process of the Joint-AI model.
Intervention: The assistance of the interpretable Joint-AI model (Diagnostic Test)
Outcomes
Primary Outcomes
Rate of correct diagnostic classification with assistance of the Joint-AI Model
Time Frame: Through study completion, an average of 1 year
The rate of correct diagnoses in discriminating pancreatic cancer from other non-cancer lesions, determined by comparing endoscopist diagnosis assisted by the Joint-AI model against the final histopathological diagnosis (reference standard).
Rate of correct diagnostic classification with assistance of the Interpretable Joint-AI Model
Time Frame: Through study completion, an average of 1 year
The rate of correct diagnoses in discriminating pancreatic cancer from other non-cancer lesions, determined by comparing endoscopist assessments assisted by the Interpretable Joint-AI model against the final histopathological diagnosis (reference standard)
Secondary Outcomes
- Rate of correct diagnostic classification of the Joint-AI model and the interpretable Joint-AI model(Through study completion, an average of 1 year)
- Endoscopist-reported confidence score in diagnosis with AI assistance (the score is on a scale of 0%-100%, where 0 represents "not confident at all" and 100 represents "completely confident")(Through study completion, an average of 1 year)
- Rate of correct diagnostic classification of endoscopists without AI assistance(Through study completion, an average of 1 year)
Investigators
Bin Cheng
Professor
Huazhong University of Science and Technology