Edema Estimation From Facial Images Taken Before and After Dialysis via Contrastive Multi-Patient Pre-Training

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Edema is a common complication in kidney disease patients.
• Accurate, quantitative measurement of edema is crucial for patient management.
• Current methods for edema assessment may be subjective or invasive.

Purpose of the Study:

• To develop and evaluate a novel method for estimating the degree of edema using facial images.
• To improve the accuracy of pre- and post-dialysis classification and body weight prediction in renal failure patients.
• To introduce a multi-patient pre-training framework with a novel contrastive learning approach for edema estimation.

Main Methods:

• Development of a multi-patient pre-training framework for edema knowledge acquisition.
• Proposal of a novel contrastive representation learning technique: weight-aware supervised momentum contrast (WeightSupMoCo).
• Application of the pre-trained model for pre- and post-dialysis classification and body weight prediction using facial images.

Main Results:

• The WeightSupMoCo approach enhances feature representations based on patient weight similarity.
• Pre-training significantly improved pre- and post-dialysis classification accuracy by 15.1%.
• Mean absolute error for body weight prediction was reduced by 0.243 kg compared to training from scratch.

Conclusions:

• The proposed method accurately estimates edema degree from facial images.
• The AI-driven edema estimation system shows potential benefits for dialysis patient care.
• Facial image analysis offers a non-invasive approach for monitoring edema in renal failure.