Enhanced Assessment of the Wound-Healing Process by Accurate Multiview Tissue Classification

Wound image evaluation with machine learning

“…Our segmentation strategy presented in Section 2.1 has some important advantages with respect to other wound image segmentation approaches used in similar studies [14,12]. On one hand, image resolution does not need to be reduced (as it has to in [12], for example) to have acceptable processing times.…”

“…Although previous studies have highlighted the adequacy of diverse ML approaches for classifying PU tissues from wound segmented images [11,44,[12][13][14]-with some authors proposing complex hybrid ML architectures and heuristic models to deal with effective tissue recognition [12]-most of these works lack an exhaustive analysis that addresses robust model fitting. In this study, parameter fitting has been supported by 10-fold cross-validation analysis on a training set, and a rigorous statistical point of view has been followed (using statistical tests, p-values and confidence intervals) to measure the significance of the differences in efficacy obtained with the different ML models and parameters.…”

“…One of these aspects involves the partial problem of wound area identification, which has been tackled with different techniques such as contour detection with histogram segmentation, active contours modelling, region growing, clustering approaches or skin texture models [8][9][10]. Unlike these proposals, other approaches focus on detecting the different tissues existing in the wound, by using diverse segmentation methods-such as histogram thresholding, watersheds, mean-shift smoothing, region growing, classification or graphs-sometimes combined with machine learning (ML) strategies [11][12][13][14].…”

“…In [21] Wannous et al used three different methods for region segmentation as well as a SVM to classify tissues in images of pressure sores and ulcers, but their approach required a preliminary manual selection of the area of interest in the wound; moreover, their results show high classification error scores, which are even greater than 50% for some critical tissues such as necrosis (see [21], table IV). More recently, Wannous's team has published [14] an innovative approach that combines 3-D wound surface measurements with tissue classification based on a SVM model to achieve enhanced wound healing assessments. Finally, in [12] we presented a mean-shift procedure along with a region-growing strategy for effective region segmentation, combined with a complex hybrid approach based on acascade of single-class classifiers for wound tissue recognition.…”

Wound Healing Assessment Using Digital Photography: A Review

“…In clinical arena, assessment methods have mainly been based on Significant advances in the field of digital image processing make the standardized digital photography as the most popular tool which considers both aspects of the wound for assessment of healing. Several successful attempts to classify wound tissue have been made through segmentation algorithms applied to various sets of features extracted from different colour spaces, with the aim of objectively monitoring the colour changes during healing [8][9][10][11][12][13][14][15] and automatically determining the wound volume and area [16][17][18]. All these proposed methods for assessing wound healing provide information only at the surface level and cannot be applied to all types of wound [2,19].…”