Establishing a machine learning model for predicting nutritional risk through facial feature recognition.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Nutritional Science

Background:

• Malnutrition is a global health concern requiring effective nutritional risk assessment.
• Current methods may be invasive or time-consuming.
• Early identification and intervention are crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning model for predicting nutritional risk using facial feature recognition.
• To establish a non-invasive and efficient method for early nutritional status evaluation.

Main Methods:

• Utilized medical examination data and facial images from 949 patients.
• Employed U-net for orbital fat pad segmentation and Histogram of Oriented Gradients (HOG) for feature extraction.
• Applied Principal Component Analysis (PCA) for dimensionality reduction and Support Vector Machine (SVM) for NRS-2002 risk score prediction.

Main Results:

• Achieved a 73.1% accuracy in predicting NRS-2002 nutritional risk scores.
• Demonstrated higher accuracy in the elderly group (85%) compared to the non-elderly (71.1%).
• Showcased varying accuracies across genders and hospital locations, with remote areas yielding 76.5%.

Conclusions:

• The facial recognition model shows significant potential for feasible and accessible nutritional risk assessment.
• The 73.1% accuracy indicates a viable non-invasive tool for early detection and intervention.
• Further improvements could enhance this innovative approach for widespread clinical application.