Development of Infection Category Detection in Diabetic Foot Ulcers Using Machine Learning

Abstract

Introduction: Traditional infection detection methods, including clinical examinations and microbiological cultures, are time-consuming. Artificial Intelligence (AI), particularly machine learning and deep learning, offers a promising alternative by swiftly and accurately analyzing medical data.

Objectives: This study aimed to identify the role of AI in detecting infections in DFUs using image recognition algorithms and prediction models based on patient clinical parameters.

Methods: An image dataset of DFU infections, categorized into three classes (mild, moderate, and severe), was utilized, consisting of 90 images (30 images per class). Features were extracted using the Gray Level Co-occurrence Matrix (GLCM) from segmented images. To handle non-linear data, the Support Vector Machine (SVM) algorithm was employed with a kernel trick, mapping lower-dimensional data to higher dimensions to achieve linear separability. The model's performance was evaluated using accuracy, precision, and recall.

Results: The SVM classifier achieved an overall accuracy of 71.4%. It performed exceptionally for moderate infections, achieving perfect precision, recall, and F1-score of 1.00. For mild infections, the precision was 0.50, recall was 1.00, and F1-score was 0.67, indicating some misclassifications. Severe infections had a precision of 1.00, recall of 0.67, and F1-score of 0.80, suggesting a conservative approach that missed some severe cases.

Conclusions: Future improvements could include hyperparameter tuning, expanding training datasets, and employing advanced techniques like SMOTE or ADASYN for class imbalance. Despite current limitations, AI shows significant potential in revolutionizing diabetic wound management, offering more timely and effective infection detection, thus improving clinical outcomes.