Estimation of Wound Area and Severity Level of Skin tear using Deep Learning Methods*

Skin tears occur mainly in older adults, making it difficult to identify the wound area and severity level when making care decision. We propose an algorithm for estimating the wound area and severity level of skin tears using a deep learning method. In this study, U-Net was used to estimate the skin tear area and VGG16 was used to estimate the severity level. The deep learning method shows an Intersection of Union (IoU) of 0.58 and 0.65 in estimating wound areas and purpura areas, and 62.2% accuracy in estimating severity levels. The proposed method outperforms the previous method using a classical machine learning method. This indicates that the proposed deep learning method is promising for image processing for skin tears, even if the skin tears include narrow wound edges and flaps, which are difficult to distinguish from the wound area.Clinical relevance–The proposed method can automatically estimate the area and severity level of skin tears to assist caregivers who are unfamiliar with skin tears.