Can machine learning be used to discriminate between burns and pressure ulcer?

Abstract

More than 10 million people requiring medical attention and 265,000 deaths are recorded every year in the world as a result of burn injuries in both low/middle- and high-income countries. Patients with acute burn injuries are at risk of developing hospital-acquired complications due to long hospitalizations such as bedsore (pressure ulcer) as a result of immobility. The developed complications are wounds that share similar physical appearances with burns wounds, as such misdiagnosis may likely to occur and leads to long hospitalization and increases cost of medical intervention. This study focused on discriminating burns and pressure ulcer using machine learning approach. We used transfer learning technique where pre-trained deep Convolutional Neural Networks (ConvNet) which includes VGG-face model and two variants of Residual Network (ResNet101 and ResNet152) were employed for extracting discriminatory features from the images and subsequently these features were fed into a Support Vector Machine for classification. Our result shows a recognition accuracy of up to 99.9%.