Convolutional neural network architectural models for multiclass classification of aesthetic facial skin disorders

Abstract

Aesthetic facial skin disorders present a significant challenge in dermatology, often requiring accurate diagnosis for effective treatment. This study investigates the effectiveness of convolutional neural network (CNN) architectural models for multiclass classification of facial skin disorders, including oily skin, hyperpigmentation, acne, redness, blackheads, and normal skin types. The research aims to compare the performance of various CNN architectures, including EfficientNetB0, HRNet, DenseNet201, and ResNet50, in classifying facial skin image datasets into multiple categories of aesthetic skin disorders. Additionally, this study examines the impact of feature combination methods on classification accuracy by enhancing attribute representation. Specifically, it explores the integration of color moment (CM) for color features and Laplacian of Gaussian (LoG) for shape features with CNN architectural models. Performance evaluation and comparison are conducted between CNN architectures with and without feature combinations. The results demonstrate that the selected CNN architectures effectively classify various aesthetic facial skin disorders, achieving an accuracy rate of 0.95 using the ResNet50 architecture. Moreover, the findings highlight that feature combination techniques, particularly the integration of CNN architectures with CM and LoG, significantly enhance classification accuracy. This study emphasizes the potential of CNN architectural models to improve early diagnostic capabilities in dermatology. The integration of artificial intelligence (AI)-based technology in the classification of aesthetic facial skin disorders offers a promising avenue for more accurate and effective diagnosis and treatment, ultimately improving patient outcomes.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint