Infrared Thermal Images Classification for Pressure Injury Prevention Incorporating the Convolutional Neural Networks

Hospital-acquired pressure injury is difficult to identify in the early stage, accompanied with increased morbidity but considered to be preventable. For helping the nurses to monitor the status of the patients' skin, the infrared thermal imaging and the convolutional neural networks were integrated to identify and prevent pressure injury. In the first stage, infrared thermal images were shoot and labelled with the normal group and the pressure injury group by the clinical nurses. In the second stage, the convolutional neural networks and two machine learning algorithms, the random forest and the support vector machine, were applied to classify these two classes of the collected images. The classification model was trained on 164 images and was tested on the special image dataset consisted of 82 infrared thermal images of 1 day before pressure injury. Gray level co-occurrence matrix was utilized to extract the texture features of the infrared thermal images and we chose the pearson correlation coefficient and the Chi square test as the feature selection methods. The classification accuracy of the proposed convolutional neural networks model was 95.2% and the area under curve was 0.98. Moreover, the classification results from the test dataset were conformed to the experience of the experts. After feature selection, variance and entropy were proved to the best distinguishable features. Finally, we concluded that combining the infrared thermal imaging and convolutional neural networks could contribute to the prevention of pressure injury. This measure should be performed in high-risk populations to reduce the incidence of pressure injury.