Roberto Mazzetto
Alvise Sernicola
Jacopo Tartaglia
Christian Ciolfi
Mauro AlaibacIntroduction
Vitiligo is a common, acquired, patchy depigmented skin disease that can be localized or widespread. Its global prevalence ranges from 0.2% to 1.8% (1). Despite its benign nature, the psychosocial burden of vitiligo parallels that of other chronic dermatological diseases such as psoriasis and eczema (2). Lesions, particularly those in visible areas like the face, neck, and hands, contribute to stigma, anxiety, and depression. The chronic nature of vitiligo, combined with the reluctance of many dermatologists to offer active treatment due to pessimism regarding its efficacy, significantly affects patients' quality of life (3). Traditional diagnostic tools such as dermatoscopy, Wood's lamp examination, and skin biopsies continue to play a central role. Dermatoscopy provides detailed insights into pigment changes and structural alterations in vitiligo patches. Wood's lamp is helpful in delineating lesions and assessing disease stability, especially in subjects with light phototypes. Conversely, biopsies are invasive and therefore should be reserved for select cases. Assessing vitiligo disease severity and progression remains challenging due to variability in lesion size, number, distribution, morphology, and pigmentation patterns. In fact, several subjective and semi-objective tools, including the Vitiligo European Task Force assessment (VETFa) and the Vitiligo Area Scoring Index (VASI), have been developed (4, 5), but they are limited by lack of objectivity, consistency, and ease of use in everyday clinical settings (6). Objective methods, such as colorimetry, reflectance confocal microscopy (RCM), and digital image analysis, are promising but often require expensive equipment and specialized expertise (7). Recently, AI and deep learning methods have emerged as powerful tools in medical image analysis, enabling automated pattern recognition, classification, prediction, and decision support. These tools enable objective and reproducible assessment of vitiligo, thereby assisting clinicians in evaluating treatment efficacy and tailoring therapy within a personalized medicine approach.
In this article, we review the principal AI methods currently available to assist clinicians in diagnosing and monitoring vitiligo, evaluating how these tools can integrate into and enhance current clinical practice.
Various machine learning (ML) approaches have been utilized to distinguish between healthy individuals and vitiligo patients and differentiate segmental from non-segmental forms. Recent AI-driven models have enhanced diagnostic capabilities. Hypopigmented dermatoses (HD), a group encompassing vitiligo, pityriasis alba, pityriasis versicolor, and others, pose diagnostic challenges (8). Deep learning has demonstrated higher accuracy rates for vitiligo than for other HD types, attributable to greater sample sizes and better feature learning (9). Studies such as those by Han et al. (10), which achieved 90% accuracy in classifying 12 skin diseases, and Esteva et al. (11), where AI models matched dermatologist-level accuracy in skin cancer detection, highlight their strong diagnostic potential in dermatology, particularly for image-based disease recognition and triage support. Severity evaluation, however, remains crucial for treatment planning. Although VASI is widely used, manual calculation is complicated by irregular lesion shapes. Image analysis software like Photoshop® and AutoCAD® offers improved area measurements but demands significant time and expertise (12). AI models now enable automatic lesion segmentation and colorimetric analysis, streamlining severity assessment (9).
Several deep learning models have been developed for vitiligo analysis with high accuracy and clinical relevance. Convolutional Neural Networks (CNNs) are deep learning architectures designed to automatically learn spatial hierarchies of features from input images. EfficientNet B7-based CNNs have shown strong classification performance, supporting preliminary assessments of vitiligo images (13). Other new CNN-based approaches, such as SE_ResNet-18 and Swin Transformer Large models, have achieved accuracies exceeding 93% (14). Class activation maps generated by these models enhance interpretability, linking model decisions to lesion characteristics and facilitating clinical trust. AU-Net (Attention U-Net), an enhanced AI architecture with integrated attention mechanisms, achieved an estimated accuracy of over 90% in segmenting depigmented areas by selectively focusing on relevant spatial features while reducing background interference (15). EfficientNetV2-L, a CNN optimized for both speed and accuracy, reached 94.5% classification accuracy and demonstrated strong generalizability through cross-validation and external clinical testing (16). AI-based models have also demonstrated high accuracy in vitiligo severity assessment by replicating dermatologist-assigned scores, offering a reliable and objective alternative for clinical evaluation (17). Advanced hybrid AI models using architectures such as YOLO v3 and UNet++ have also achieved reliable morphometric and colorimetric assessments in patients with Fitzpatrick skin types III-IV. Add-on metrics (VAreaA, VAreaR, VColor) correlated well with dermatologist evaluations, making these models suitable for diverse clinical settings (18). In addition to clinical validation, an important consideration is “segmentation,” which refers to the process of partitioning an image into meaningful regions, typically by classifying each pixel, based on visual characteristics such as color, texture, or edges, to isolate and analyze specific structures or objects. Hybrid ViT-CNN leverages the global context modeling of Vision Transformers and the fine-grained spatial details, achieving 96.8% segmentation accuracy (16). Finally, VitiligoNet, a deep CNN framework, offers end-to-end lesion detection, segmentation, and classification, with a reported accuracy of 97.4% that has proven effective across diverse skin tones and clinical scenarios (19). Table 1 compares AI models for vitiligo diagnosis and severity assessment.
Table 1. Comparison of AI models for vitiligo diagnosis and severity assessment.
Discussion
Although dermatology has increasingly embraced AI for various tasks (such as melanoma detection, lesion classification, risk stratification, and dermoscopic image analysis) vitiligo-specific applications are still emerging and progressively expanding (20). Prospective AI developments include training CNN models on images of vitiligo alongside other HDs to improve lesion detection and classification. Such tools could replace Wood's lamp analysis when unavailable. In addition, automated VASI scoring would facilitate remote monitoring, empowering patients to track disease progression and reducing clinical burdens.
Recent research highlights AI's role in predictive analytics. For instance, elevated full blood count parameters (B lymphocyte count and natural killer cell count) could correlate with localized vitiligo, while systemic markers could predict non-segmental disease progression with promising performance (accuracy 73%) (21). This integrative approach is also useful to identify key disease pathways, such as p38 MAPK signaling and oxidative stress modulation, underscoring AI's dual utility in clinical severity scoring and therapeutic target discovery for vitiligo (18, 22). Interpretability remains crucial but visualizing feature maps from neural networks offers insights into internal processing, promoting transparency and fostering clinician confidence. Therefore, AI models could not only provide disease classification labels but also enhance clinical decision-making by offering intuitive, image-based insights that can simplify and complement traditional assessment methods based on clinical variables and scoring systems.
Despite notable progress, most current studies rely on limited datasets and broader validation through multicenter, multiethnic cohorts is needed to overcome this limitation (14). Moreover, models must handle small lesions, subtle repigmentation, and heterogeneous skin tones to ensure robust performance. Ongoing research should focus on larger datasets, multimodal analyses (combining clinical images, laboratory data, and patient-reported outcomes), and real-world validation. Importantly, dermatologists must be actively involved in shaping the ethical and practical frameworks governing AI applications.
Accurate objective measurement is important to classify and standardize the effectiveness of therapeutic agents, especially in the clinical trial setting of new drugs. However, it is known that objective measures provide an incomplete assessment of severity and response to treatment and that subjective patient-reported measures need to be included. For this reason, AI models should not only consider the accuracy of the data collection method but should be trained to integrate the “human” component of daily life and clinical experience which relies on the subjective patient and clinician point of view. This will be particularly useful in the clinical practice to compare the efficacy between treatment options and to tailor the preferred option according to each patient (23).
In fact, even if progress may be slow, disease stabilization and repigmentation are achievable goals, especially when there is a shared understanding between patients and healthcare providers regarding vitiligo and its treatment options (24). The psychological burden associated with the condition is in fact described in various papers that highlight the importance of raising awareness among both physicians and patients about the well-documented relationship between vitiligo and anxiety or other stress-related emotions (25, 26).
The journey toward routine application in everyday clinical settings is restricted by the following issues: regulatory approval under medical device regulations is necessary to allow marketing of image analysis software; compliance to data privacy laws must guarantee informed patient consent and secure use of clinical images that constitute sensitive personal health information; medico-legal concerns related to healthcare provider responsibility, role of clinicians in mitigating the risk of errors and validation in different patient populations still need to be adequately addressed. Finally, practical aspects related to image analysis, such as the scarce or variable contrast that may be associated with low Fitzpatrick phototypes or with seasonal changes in skin color, respectively, may complicate automated interpretation of patient images.
AI is anticipated to be a transformative force in dermatology: from improved diagnostic accuracy to personalized treatment monitoring, AI tools can enhance clinical workflows and patient experiences within a human-centered healthcare model.
With new therapies on the horizon and AI-driven diagnostics becoming a reality, vitiligo is poised to emerge from obscurity into a new era of personalized, technology-enhanced care.
Author contributions
RM: Writing – review & editing, Writing – original draft. AS: Writing – review & editing, Supervision, Writing – original draft, Conceptualization. JT: Writing – review & editing. CC: Writing – review & editing. MA: Writing – review & editing.
Funding
The author(s) declare that financial support was received for the research and/or publication of this article. Open Access funding provided by Università degli Studi di Padova | University of Padua, Open Science Committee.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declare that Gen AI was used in the creation of this manuscript. Generative AI technology was used to assist in the editing of written content in this article. ChatGPT version GPT-4, model “gpt-4o” (OpenAI, San Francisco, CA, USA) was employed via the ChatGPT platform (https://chat.openai.com). The authors have reviewed and verified the accuracy and originality of all AI-assisted content.
Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1. Zhang Y, Cai Y, Shi M, Jiang S, Cui S, Wu Y, et al. The prevalence of vitiligo: a meta-analysis. Oiso N, ed. PLoS ONE. (2016) 11:e0163806. doi: 10.1371/journal.pone.0163806
2. Linthorst Homan MW, Spuls PI, De Korte J, Bos JD, Sprangers MA, Van Der Veen JPW, et al. The burden of vitiligo: patient characteristics associated with quality of life. J Am Acad Dermatol. (2009) 61:411–20. doi: 10.1016/j.jaad.2009.03.022
3. Salzes C, Abadie S, Seneschal J, Whitton M, Meurant J-M, Jouary T, et al. The vitiligo impact patient scale (VIPs): development and validation of a vitiligo burden assessment tool. J Investig Dermatol. (2016) 136:52–8. doi: 10.1038/JID.2015.398
4. Taïeb A, Picardo M, on behalf of the other VETF members. The definition and assessment of vitiligo: a consensus report of the Vitiligo European Task Force. Pigment Cell Res. (2007) 20:27–35. doi: 10.1111/j.1600-0749.2006.00355.x
5. Hamzavi I, Jain H, McLean D, Shapiro J, Zeng H, Lui H. Parametric modeling of narrowband UV-B phototherapy for vitiligo using a novel quantitative tool: the Vitiligo Area Scoring Index. Arch Dermatol. (2004) 140:677–83. doi: 10.1001/archderm.140.6.677
6. Mikhael N, Sabry H, El-Refaey A, Salem R, El-Gendy M, Farid S. Clinimetric analysis of recently applied quantitative tools in evaluation of vitiligo treatment. Indian J Dermatol Venereol Leprol. (2019) 85:466. doi: 10.4103/ijdvl.IJDVL_63_17
7. Kang HY, Bahadoran P, Ortonne J. Reflectance confocal microscopy for pigmentary disorders. Exp Dermatol. (2010) 19:233–9. doi: 10.1111/j.1600-0625.2009.00983.x
8. Tanvir S, Syed SA, Hussain S, Zia R, Rashid M, Zahid H. Detection of vitiligo through machine learning and computer-aided techniques: a systematic review. Shah S, ed. BioMed Res Int. (2024) 2024:3277546. doi: 10.1155/bmri/3277546
9. Huang H. Intelligent diagnosis of hypopigmented dermatoses and intelligent evaluation of vitiligo severity on the basis of deep learning. Dermatol Ther. (2024) 14:3307–20. doi: 10.1007/s13555-024-01296-9
10. Han SS, Kim MS, Lim W, Park GH, Park I, Chang SE. Classification of the clinical images for benign and malignant cutaneous tumors using a deep learning algorithm. J Invest Dermatol. (2018) 138:1529–38. doi: 10.1016/j.jid.2018.01.028
11. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. (2017) 542:115–8. doi: 10.1038/nature21056
12. Pellacani G, Farnetani F, Ciardo S, Chester J, Kaleci S, Mazzoni L, et al. Effect of reflectance confocal microscopy for suspect lesions on diagnostic accuracy in melanoma: a randomized clinical trial. JAMA Dermatol. (2022) 158:754–61. doi: 10.1001/jamadermatol.2022.1570
13. Gessert N, Nielsen M, Shaikh M, Werner R, Schlaefer A. Skin lesion classification using ensembles of multi-resolution EfficientNets with meta data. MethodsX. (2020) 7:100864. doi: 10.1016/j.mex.2020.100864
14. Zhong F. Optimizing vitiligo diagnosis with ResNet and Swin transformer deep learning models: a study on performance and interpretability. Sci Rep. (2024) 14:9127. doi: 10.1038/s41598-024-59436-2
15. Lu LY, Chen YT, Chen IL, Shih YC, Liu RTL, Lai YJ, et al. Feasibility of high-cellular-resolution full-field, artificial-intelligence-assisted, real-time optical coherence tomography in the evaluation of vitiligo: a prospective longitudinal follow-up study. Bioengineering. (2024) 11:196. doi: 10.3390/bioengineering11020196
16. Zhang X, Balasubramaniyan D, Chidambaram N, Ravi V. Predictive analysis of vitiligo treatment drugs using degree and neighborhood degree-based topological descriptors. Sci Rep. (2025) 15:5218. doi: 10.1038/s41598-025-89603-y
17. Li Pomi F, Papa V, Borgia F, Vaccaro M, Pioggia G, Gangemi S. Artificial intelligence: a snapshot of its application in chronic inflammatory and autoimmune skin diseases. Life. (2024) 14:516. doi: 10.3390/life14040516
18. Guo L, Yang Y, Ding H, Zheng H, Yang H, Xie J, et al. A deep learning-based hybrid artificial intelligence model for the detection and severity assessment of vitiligo lesions. Ann Transl Med. (2022) 10:590. doi: 10.21037/atm-22-1738
19. Zhang L, Mishra S, Zhang T, Zhang Y, Zhang D, Lv Y, et al. Design and assessment of convolutional neural network based methods for vitiligo diagnosis. Front Med. (2021) 8:754202. doi: 10.3389/fmed.2021.754202
20. Haenssle HA, Fink C, Schneiderbauer R, Toberer F, Buhl T, Blum A, et al. Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol. (2018) 29:1836–42. doi: 10.1093/annonc/mdy166
21. Wang Z, Xue Y, Liu Z, Wang C, Xiong K, Lin K, et al. OPEN AI fusion of multisource data. Sci Rep. (2024) 14:24278. doi: 10.1038/s41598-024-75062-4
22. Hillmer D, Merhi R, Boniface K, Taieb A, Barnetche T, Seneschal J, et al. Evaluation of facial vitiligo severity with a mixed clinical and artificial intelligence approach. J Investig Dermatol. (2024) 144:351–7.e4. doi: 10.1016/j.jid.2023.07.014
23. Renert-Yuval Y, Desai S, Huang V, Walsh S, Rosmarin D, Silverberg N. A review of the vitiligo literature to standardize expression of disease severity. J Drugs Dermatol. (2024) 23:842–6. doi: 10.36849/JDD.8049
24. Prajapati VH, Lui H, Miller-Monthrope Y, Ringuit L, Turchin I, Hong HC-H, et al. Canadian consensus guidelines for the management of vitiligo. Dermatol Ther. (2025) 15:1351–69. doi: 10.1007/s13555-025-01402-5
25. Chen Y, Xie Y, Wu N, Li B, Tang S, Xu M, et al. Heterogeneity of anxiety in vitiligo: a systematic review and quantitative analysis with a focus on adolescents. J Psychosom Res. (2025) 195:112182. doi: 10.1016/j.jpsychores.2025.112182
Keywords: vitiligo, dermatology, AI, deep learning, image analysis
Citation: Mazzetto R, Sernicola A, Tartaglia J, Ciolfi C and Alaibac M (2025) Potential of automated image analysis for the measurement of vitiligo lesions. Front. Med. 12:1623408. doi: 10.3389/fmed.2025.1623408
Received: 05 May 2025; Accepted: 24 July 2025;
Published: 14 August 2025.
Edited by:
Yan Valle, Vitiligo Research Foundation, United StatesReviewed by:
Priyansh Gupta, ESIC Medical College (Faridabad), IndiaCopyright © 2025 Mazzetto, Sernicola, Tartaglia, Ciolfi and Alaibac. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Alvise Sernicola, YWx2aXNlLnNlcm5pY29sYUB1bmlwZC5pdA==