Classification of pressure ulcers using machine learning techniques

Pressure ulcers are wounds caused by prolonged inactivity in bedridden patients. This situation has become an important health problem in the world. Correct diagnosis of pressure ulcers is essential for treatment to be effective. Wound characteristics have an important effect on healing. Interventional methods of obtaining information are painful for patients in diagnosing pressure ulcers. In addition, these methods may cause the patient to become infected. Therefore, non-surgical wound tracing techniques should be preferred. With imaging systems, it is ensured that the characteristics of the wound are analyzed accurately without contacting the wound. The aim of this thesis study is to make a positive contribution to the treatment processes or to prevent the formation of wounds with the classification of pressure ulcers by using machine learning techniques in image analysis. In this thesis, an innovation has been brought to the pressure sore problem in the literature in terms of the number of staging. Pressure ulcer staging is handled as an accurate image classification problem. Real hospital data consisting of 697 images wound Logistic Regression, Neural Networks and Support Vector Machines were analyzed by the method. Features such as wound color and size in these images were separated by image processing and the stage of the wound was determined from the images. The 6 stages of pressure ulcers are referenced for classification. Although it is not difficult for experts to determine the class values for classification, the images obtained from different angles in a small number of data sets in this study make it difficult to classify the images in the background. Using more data for training and using Deep Learning architectures will increase performance values.