Abstract
BACKGROUND: Lower limb edema is a common clinical symptom closely associated with chronic diseases such as heart failure, liver disease, and renal dysfunction. Edema severity grading is an important indicator for clinical diagnosis and disease monitoring. However, traditional assessments that rely on visual inspection and manual palpation are subjective and inconsistent, making them insufficient to meet the requirements of standardized and precise diagnostics. METHODS: This study proposes a multistage deep learning framework that integrates object detection and image classification for automatic detection and grading of lower limb edema. The system architecture initially employed YOLO models to detect indentation regions, followed by image enhancement techniques to improve the representation of edema features and enhance the detection accuracy. Finally, classification models were used to categorize edema severity. To address the data imbalance issue, random rotation was applied for data augmentation, and non-target regions were removed through automatic background elimination and cropping to enhance classification performance. RESULTS: The experimental findings demonstrated that the proposed system achieved an average classification accuracy of 87~93% across different edema severity levels, 90-94% for recall rates, and 93~97% for precisions for different edema stages. These results validate the feasibility and effectiveness of the automatic detection and grading classification system for lower limb edema. CONCLUSION: The proposed system holds the potential for both clinical decision support and home-based self-care, enhancing the accuracy and consistency of edema assessment for patients and healthcare professionals. It can facilitate smart and precision medicine based on the status of lower limb edema.