Video-Based Automatic Quantification of Leg Edema: a Pilot Study in Patients With Hemodialysis With and Without Heart...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Clinical Diagnostics

Background:

• Pitting edema assessment is subjective and challenging, particularly in remote healthcare scenarios.
• Objective and reliable methods for grading edema severity are needed to improve patient management.

Purpose of the Study:

• To develop and evaluate a video-based deep learning (DL) model for objective classification of pitting edema severity.
• To address the subjectivity inherent in traditional edema assessment methods.

Main Methods:

• A convolutional neural network (EfficientNetB0) was trained on 247 videos of pretibial images from 34 hemodialysis patients.
• The model was trained to classify pitting edema on a 0-4 scale using pre- and post-pressing images.

Main Results:

• The DL model achieved 81.5% accuracy, 81.2% sensitivity, and 81.9% specificity in distinguishing severe edema (grades 3-4) from mild edema (grades 0-1).
• Accuracy reached 85.8% when differentiating extreme cases (grade 0 vs. grade 4).

Conclusions:

• This pilot study demonstrates the feasibility of using video-based DL for objective pitting edema detection.
• Further research with larger, diverse datasets and clinical validation is required for widespread implementation and generalization of the model.