Sodium Intake Estimation in Hospital Patients Using AI-Based Imaging: Prospective Pilot Study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nutritional Science

Background:

• Accurate measurement of sodium intake is crucial for hospitalized patients.
• Traditional methods can be burdensome and less precise.
• Artificial intelligence (AI)-based imaging offers a novel approach to dietary assessment.

Purpose of the Study:

• To evaluate the applicability of an AI-powered diet management system for assessing sodium content in hospital meals.
• To determine the accuracy of AI-based imaging in estimating patient sodium intake compared to established methods.

Main Methods:

• A hybrid AI model, incorporating You Only Look Once version 4 (YOLOv4) and ResNet-101, analyzed food images taken before and after meals.
• Hyperspectral imaging techniques were employed for precise food quantity and sodium amount estimation.
• 24-hour urine sodium (UNa) was used as the reference standard for sodium intake evaluation.

Main Results:

• The AI algorithm (AI-Na) estimated a median daily sodium intake of 2022.7 mg per person.
• A significant correlation was found between AI-Na and 24-hour UNa, despite some disparities.
• Regression analysis identified patient factors influencing the AI-Na and 24-hour UNa relationship, leading to a predictive formula.

Conclusions:

• AI-based imaging demonstrates clinical significance for estimating sodium intake in hospitalized patients.
• This AI approach may obviate the need for 24-hour urine sodium measurements.
• The accuracy of AI-derived sodium intake can be influenced by factors like diuretic use.