A superpixel-driven deep learning approach for the analysis of dermatological wounds

Abstract

Background. The image-based identification of distinct tissues within dermatological wounds enhances patients' care since it requires no intrusive evaluations. This manuscript presents an approach, we named QTDU , that combines deep learning models with superpixel-driven segmentation methods for assessing the quality of tissues from dermatological ulcers. Method. QTDU consists of a three-stage pipeline for the obtaining of ulcer segmentation, tissues' labeling, and wounded area quantification. We set up our approach by using a real and annotated set of dermatological ulcers for training several deep learning models to the identification of ulcered superpixels. Results. Empirical evaluations on 179,572 superpixels divided into four classes showed QTDU accurately spot wounded tissues (AUC = 0.986, sensitivity = 0.97, and specificity = 0.974) and outperformed machinelearning approaches in up to 8.2% regarding F1-Score through fine-tuning of a ResNet-based model. Last, but not least, experimental evaluations also showed QTDU correctly quantified wounded tissue areas within a 0.089 Mean Absolute Error ratio. Conclusions. Results indicate QTDU effectiveness for both tissue segmentation and wounded area quantification tasks. When compared to existing machine-learning approaches, the combination of superpixels and deep learning models outperformed the competitors within strong significant levels. can be automatically evaluated by Computer-Aided Diagnosis (CAD) tools, or even used for the searching of massive databases through content-only queries, as in Content-Based Image Retrieval (CBIR) applications. In both CAD and CBIR cases, the detection of abnormalities requires the extraction of patterns from images, while a decision-making strategy is necessary for juxtaposing new images to those in the database [4,5].Since dermatological lesions are routinely diagnosed by biopsies and surrounding skin aspects, ulcers can be computationally characterized by particular types of tissues (and their areas) within the wounded region [6,7]. For instance, Mukherjee et al.[8] proposed a five-color classification model and applied a color-based low-level extractor further labeled by a Support-Vector Machine (SVM) strategy at an 87.61% hit ratio. Such idea of concatenating feature extraction and classification is found at the core of most wound segmentation strategies, as in the study of Kavitha et al.[9] that evaluated leg ulcerations by extracting patterns based on local spectral histograms to be labeled by a Multi-Layer Perceptron (MLP) classifier with 87.05% accuracy. Analogously, Pereyra et al.[10] discussed the use of color descriptors and an Instancebased Learning (IbL) classifier with a 61.7% hit ratio, whereas Veredas et al. [11] suggested the use of texture descriptors and an MLP classifier with 84.84% accuracy.Blanco et al.[4] and Chino et al.[12] followed a slightly different premise for finding proper similarity measures and comparison criteria for dermatological wounds. Their approaches are based on a divide-andconquer stra...

Superpixel Image Segmentation of the Tumoral Microenvironment in Colorectal Cancer: Frontiers Beyond the Microscope

“…Superpixel methods [18] are gaining traction in the eld of medicine and recent attempts have been made to combine SIS and DL. In dermatology the combination of superpixels and deep learning models outperformed other competing methods [19]. The advantages for SIS in the deep learning space is that is can represent the structure of an image in adaptive sizes and shapes, with the ability to improve classi cation performance, especially for noisy classi cation (corrupted labels), as well as boundary misclassi cation [20].…”

Superpixel Image Segmentation of the Tumoral Microenvironment in Colorectal Cancer: Frontiers Beyond the Microscope

“…to combine SIS and DL. In dermatology the combination of superpixels and deep learning models outperformed other competing methods [19]. The advantages for SIS in the deep learning space is that is can represent the structure of an image in adaptive sizes and shapes, with the ability to improve classi cation performance, especially for noisy classi cation (corrupted labels), as well as boundary misclassi cation [20].…”

A novel approach to extracting useful information from noisy TFDs using 2D local entropy measures

“…FP and FN are the classification version of statistical error types 1 and 2. This metrics are used in several papers that deal with image [47,48] and signal processing [49], such as EEG signals [50,51]. In addition to numerical results, images of the obtained signal masks are shown of Figs.…”