Predicting Bum Injury Intensity from Skin Bum Images with Hybrid Machine Learning Technique

Abstract

Burns are one of the principal contributors of major clinical catastrophe, which can result in strong physiological reaction with long-lasting effects if patients are not appropriately treated. Predicting burn wound intensity at a preliminary phase is quite difficult, which is essential for proper burn treatment. Thus, a hybrid machine learning based approach has been proposed in this study to predict the bum injuiy intensity autonomously from skin bum images of patients. The proposed technique includes a Convolutional Neural Network(CNN) incorporating transfer learning method as the feature extractor from the input images and then conventional machine learning classification models as the classifiers for categorizing the images into three classes according to their bum severity: first, second and third degree bums. Ten different types of classification models have been trained and tested to explore the best performing one. Moreover, to validate the efficacy and potency of the proposed technique, the predictive analysis has also been carried out in another approach where digital image processing stages have been used as the feature extractor and then machine learning classification models as classifiers. According to the comparative evaluation findings, each classifiers using the suggested CNN architecture as feature extractor achieves noticeably higher accuracy and among them ‘XGBoost’ classification model outperforms others with 96.42% accuracy employing the proposed technique.