Classification of non-tumorous skin pigmentation disorders using voting based probabilistic linear discriminant analysis

Abstract

Non-tumorous skin pigmentation disorders can have a huge negative emotional impact on patients. The correct diagnosis of these disorders is essential for proper treatments to be instituted. In this paper, we present a computerized method for classifying five non-tumorous skin pigmentation disorders (i.e., freckles, lentigines, Hori's nevus, melasma and nevus of Ota) based on probabilistic linear discriminant analysis (PLDA). To address the large within-class variance problem with pigmentation images, a voting based PLDA (V-PLDA) approach is proposed. The proposed V-PLDA method is tested on a dataset that contains 150 real-world images taken from patients. It is shown that the proposed V-PLDA method obtains significantly higher classification accuracy (4% or more with p< 0.001 in the analysis of variance (ANOVA) test) than the original PLDA method, as well as several state-of-the-art image classification methods. To the authors' best knowledge, this is the first study that focuses on the non-tumorous skin pigmentation image classification problem. Therefore, this paper could provide a benchmark for subsequent research on this topic. Additionally, the proposed V-PLDA method demonstrates promising performance in clinical applications related to skin pigmentation disorders.