Using 3D facial information to predict malnutrition and consequent complications

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Phase angle (PhA) is a marker of body composition and nutritional status.
• Predicting malnutrition is crucial for patient prognosis.
• Current methods for malnutrition assessment can be invasive or time-consuming.

Purpose of the Study:

• To investigate the feasibility of predicting PhA-diagnosed malnutrition using deep learning (DL) on facial images.
• To develop and validate a DL framework for automated malnutrition assessment.
• To explore the correlation between PhA and clinical outcomes.

Main Methods:

• A multimodal DL framework was developed to analyze 3D facial data.
• The framework was trained and validated using cross-validation on inpatient data.
• Subjective global assessment served as the gold standard for malnutrition diagnosis.

Main Results:

• The DL model achieved an AUC of 0.77 and an accuracy of 0.74 in predicting PhA.
• Low PhA was associated with a higher incidence of infectious complications (P=0.003).
• The model demonstrated fair performance in predicting PhA categories from facial images.

Conclusions:

• Facial image analysis using DL is a feasible approach for predicting patient PhA.
• This non-invasive method can aid in malnutrition assessment and prognostic prediction.
• The developed DL framework offers a promising tool for objective nutritional status evaluation.