A Food Intake Estimation System Using an Artificial Intelligence-Based Model for Estimating Leftover Hospital Liquid...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Nutritional Science

Background:

• Accurate patient food intake assessment is crucial but challenging with traditional visual estimation methods in clinical settings.
• Existing methods for measuring dietary intake are often imprecise and labor-intensive, necessitating simpler, more accurate alternatives.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) based system for estimating leftover liquid food intake in hospitalized patients.
• To compare the accuracy of the AI food intake estimation system against visual estimation methods (image-based and direct) and a weighing method.

Main Methods:

• An AI model was developed to estimate the energy content of leftover liquid food from images.
• The AI's estimations were compared against dietitian image visual estimations and nurse direct visual estimations using 300 liquid food samples.
• Root-mean-square error (RMSE) and coefficient of determination (R²) were used, with Spearman rank correlation and t-tests against a weighing method for validation.

Main Results:

• The AI estimation method showed a smaller RMSE (8.12 kcal) than image visual estimation (8.49 kcal) but larger than direct visual estimation (4.34 kcal).
• AI estimation demonstrated no significant difference from actual values (P=.82), unlike image (P<.001) and direct visual estimations (P=.007).
• High correlations (ρ=0.89-0.97) were observed between AI estimations and actual values for energy, protein, fat, and carbohydrates.

Conclusions:

• The AI-based food intake estimation system accurately estimates liquid food consumption, correlating highly with the weighing method and exceeding image visual estimation accuracy.
• The AI system's errors fall within acceptable ranges, suggesting its potential applicability in clinical environments for improved dietary intake measurement.
• While promising, the AI system's accuracy is currently lower than direct visual estimation, indicating areas for further refinement.