A Pressure Sores Assessment System for Diagnosis and Decision Making: The Application of Convolutional Neuron Networks (Preprint)
BACKGROUND Pressure sores are a common problem in hospital care and long-term care. Pressure sores are caused by prolonged compression of soft tissues, which can cause local tissue damage and even lead to serious infections. Pressure sores can result in poor prognosis, long-term hospitalization, and increased medical costs, which are especially problematic in an aging society. OBJECTIVE This study uses deep learning to diagnose pressure sores and assist in making treatment decisions, thus providing additional reference for first-line caregivers. METHODS We utilized retrospective research of medical records to find photos of patients with pressure sores at National Taiwan University Hospital from 2016 to 2019. We removed the photos which were vague, underexposed, or overexposed, and then labeled the remaining photos as “infected” or “uninfected” and “extensive necrosis”, “moderate necrosis” or “limited necrosis”. Supervised machine learning was then used, and Convolutional Neural Networks (CNNs) were applied for deep learning to construct a diagnostic model. Finally, we tested the constructed model with these photos to verify its accuracy. RESULTS For the task of classification of infected and non-infected wounds, our CNN model achieved an accuracy of about 98%. For the task of classification of necrotic tissues, our model achieved accuracy of about 94%. CONCLUSIONS Compared with traditional algorithms, our deep learning model achieved higher accuracy, making it applicable in clinical circumstances.