Hexa classification of erythemato-squamous disease using deep dual features-based neural network.

Introduction: Erythemato-squamous disease (ESD) is a skin disorder characterized by both erythema and squamous changes. These conditions often involve inflammation and can manifest with various symptoms, such as itching, burning, and discomfort. However, distinguishing between different conditions solely based on external symptoms may be challenging and lead to misdiagnosis. Aim: To overcome these issues, this work proposes a novel deep learning-based Hexa-ESD framework to efficiently classify clinical skin images into hexa skin diseases. Material and methods: Initially, the clinical skin images are gathered from openly available datasets. The self-prepared clinical skin images are denoised by the Contrast stretching Adaptive Histogram Equalization (CSAHE) technique for eliminating the noise artifacts. These noise-free images are augmented with standard transformation techniques like scaling, flipping, and zooming to enhance the training dataset. The deep learning-DuoNet, which is a hybridization of DarkNet and ShuffleNet, is applied to retrieve the spatial features from the noise-free images. Then, the extracted features are fed into the walrus optimization (WalO) algorithm by dealing with complex non-linear problems for selecting the best features. These selected features are fused for classification using deep belief network (DBN) to detect the hexa ESD cases namely chronic dermatitis, lichen planus, seborrheic dermatitis, pityriasis rosea, psoriasis, and pityriasis rubra pilaris. Results: The proposed Hexa-ESD model yields an accuracy rate of 97.69% for the classification of ESD cases. From the evaluation, the Hexa-ESD framework increases the overall accuracy by 3.26%, 1.93% and 18.21% for ReliefF algorithm, Extreme Gradient Boosting and EPFS algorithm, respectively. Conclusions: The proposed Hexa-ESD framework provides an effective solution for multi-class classification of ESD using clinical skin images and reliable as a computer-aided diagnostic tool for dermatological applications.