Gastric Emptying Scintigraphy Protocol Optimization Using Machine Learning for the Detection of Delayed Gastric...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Delayed gastric emptying (GE) is a key indicator for diagnosing gastroparesis.
• Gastric emptying scintigraphy (GES) is a standard diagnostic tool for GE.
• Optimizing GES protocols can improve diagnostic efficiency.

Purpose of the Study:

• To assess the feasibility of a machine learning (ML) system for optimizing GES protocols.
• To detect delayed GE, a primary indication for gastroparesis diagnosis.
• To explore ML's potential in enhancing diagnostic accuracy and efficiency.

Main Methods:

• Developed an ML model using the JADBio AutoML AI platform.
• Trained and tested the model on 1002 patients undergoing GES.
• Utilized percent GE at various time points to predict 4-hour GE outcomes.

Main Results:

• The ML model achieved a 90.7% AUC and 80.0% balanced accuracy using 0.5-2.5 hour GE values.
• The 2.5-hour GE value alone demonstrated significant predictive power (92.4% AUC, 83.3% BA).
• The model successfully predicted 'flipping' cases with delayed GE at 4 hours.

Conclusions:

• An AI/ML model can predict delayed GE using limited imaging time points within a 4-hour GES protocol.
• This demonstrates the feasibility of using ML to optimize GES.
• ML may allow for shorter GES protocols without compromising diagnostic power.