Detection of Vitiligo Through Machine Learning and Computer-Aided Techniques: A Systematic Review

Abstract

Background and Objective: Vitiligo is a chronic skin damage disease, triggered by differential melanocyte death. Vitiligo (0.5%-1% of the population) is one of the most severe skin conditions. In general, the foundation of the condition of vitiligo remains gradual patchy loss of skin pigmentation, overlying blood, and sometimes mucus. This paper provides a systematic review of the relevant publications and conference papers based on the subject of vitiligo diagnosis and confirmation through computer-aided machine learning (ML) techniques. Materials and Methods: A search was conducted using a predetermined set of keywords across three databases, namely, Science Direct, PubMed, and IEEE Xplore. The selection process involved the application of eligibility criteria, which led to the inclusion of research published in reputable journals and conference proceedings up until June 2024. These selected papers were then subjected to full-text screening for additional analysis. Research publications that involved application of ML techniques with targeted population of vitiligo were selected for further systematic review. Results: Ten selected and screened studies are included in this systematic review after applying eligibility criteria along with inclusion and exclusion criteria applied on initial search result which was 244 studies based on vitiligo. Priority is given to those studies only which use ML techniques to perform detection and diagnosis on vitiligo-targeted population. Data analysis was carried out only from the selected and screened research articles that were published in authentic journals and conference proceedings. Conclusion: The importance of applying ML techniques in the clinical diagnosis of vitiligo can give more accurate results and at the same also eliminate the need of biased human judgement. Based on a comprehensive examination of the research, encompassing the methodologies employed and the metrics utilized to assess outcomes, it was determined that there is a need for further research and investigation regarding the application of ML algorithm for the detection and diagnosis of vitiligo with different datasets and more feature extraction.

Keywords: detection; diagnosis; image segmentation; machine learning; skin disease; systematic review; vitiligo.

Figure 1

Pie chart of no. of publications in the initial stage before applying any sort of criteria.

Figure 2

Pie chart of no. of publications included after screening and after applying exclusion and inclusion criteria.

Figure 3

PRISMA flowchart according to PRISMA guidelines and statement recommendations.

Figure 4

Scatter plot of publication year of selected and screened studies.

Figure 5

Comparison of unsupervised techniques outcomes.

Figure 6

Comparison of supervised technique outcomes.

Figure 7

Traffic light plot illustrating risk of bias.

Figure 8

Weight plot illustrating risk of bias.