Image Analysis Systems to Calculate the Surface Area of Vitiligo Lesions: a Systematic Review of Measurement Properties

Evaluation of Facial Vitiligo Severity with a Mixed Clinical and Artificial Intelligence Approach

Vitiligo is the most common depigmenting skin disorder. Given the ongoing development of new targeted therapies, it has become important to evaluate adequately the surface area involved. Assessment of vitiligo scores can be time consuming, with variations between investigators. Therefore, the aim of this study was to build an artificial intelligence system capable of assessing facial vitiligo severity. One hundred pictures of faces of patients with vitiligo were used to train and validate the artificial intelligence model. Sixty-nine additional pictures of facial vitiligo were then used as a final dataset. Three expert physicians scored the facial vitiligo on the same 69 pictures. Inter and intrarater performances were evaluated by comparing the scores between raters and artificial intelligence. Algorithm assessment achieved an accuracy of 93%. Overall, the scores reached a good agreement between vitiligo raters and the artificial intelligence model. Results demonstrate the potential of the model. It provides an objective evaluation of facial vitiligo and could become a complementary/alternative tool to human assessment in clinical practice and/or clinical research.

In-depth study of Wood's lamp examination combined with reflective confocal laser scanning microscopy for the guidance of vitiligo staging and treatment

BACKGROUND

Both Wood's lamp and reflective confidential laser scanning microcopy are helpful for the diagnosis and treatment of vitiligo. However, there is few research that contains large samples and consistent observations.

AIMS

To analyze the characteristics of Wood's lamp images and reflectance confocal microscopy (RCM) images of vitiligo lesions and to evaluate their significance in vitiligo staging.

METHODS

We analyzed the characteristics of RCM images, Wood's lamp images, the vitiligo disease activity (VIDA) score, and clinical features to guide vitiligo staging and treatment.

RESULTS

The expert consensus based on the clinical features, VIDA score, Wood's lamp findings, and isomorphic response was consistent with the Wood's lamp findings (χ2  = 3.63, p > 0.05) and RCM findings (χ2  = 3.60, p > 0.05) in diagnosing vitiligo and assessing the disease stage. There was a correlation between the three lesion grades based on the Wood's lamp findings and the stage of vitiligo (p < 0.01). Lesions that appeared porcelain white under the Wood's lamp were in the slowly progressive stage; lesions that appeared gray-white or trichromatic under the Wood's lamp were in the rapidly progressive stage; lesions with clear borders under the Wood's lamp needed further analysis by RCM for the stage to be determined; lesions with blurred borders under the Wood's lamp were in the rapidly progressive stage; lesions that were visible under the naked eye and under the Wood's lamp were in the rapidly progressive stage.

CONCLUSION

The study demonstrates a reliable correlation between the findings of RCM (a sophisticated expensive tool) and Wood's lamp examination (a simple, readily available, inexpensive tool) in the assessment of the disease activity of vitiligo lesions.

Reliability and agreement testing of a new automated measurement method to determine facial vitiligo extent using standardized ultraviolet images and a dedicated algorithm

BACKGROUND

Facial repigmentation is the primary outcome measure for most vitiligo trials. The Facial Vitiligo Area Scoring Index (F-VASI) score is often chosen as the primary outcome measure to assess the efficacy of treatments for facial vitiligo. Although useful, this scoring system remains subjective and has several limitations.

OBJECTIVES

To assess the agreement and reliability of an algorithmic method to measure the percentage depigmentation of vitiligo on the face.

METHODS

We developed a dedicated algorithm called Vitil-IA® to assess depigmentation on standardized facial ultraviolet (UV) pictures. We then conducted a cross-sectional study using the framework of the ERASE trial (NCT04843059) in 22 consecutive patients attending a tertiary care centre for vitiligo. Depigmentation was analysed before any treatment and, for 7 of them, after 3 and 6 months of narrowband UVB treatment combined with 16 mg methylprednisolone, both used twice weekly. Interoperator and interacquisition repeatability measures were assessed for the algorithm. The results of the algorithmic measurement were then compared with the F-VASI and the percentage of depigmented skin scores assessed by 13 raters, including 7 experts in the grading of vitiligo lesions.

RESULTS

Thirty-one sets of pictures were analysed with the algorithmic method. Internal validation showed excellent reproducibility, with a variation of < 3%. The percentage of depigmentation assessed by the system showed high agreement with the percentage of depigmentation assessed by raters [mean error (ME) -11.94 and mean absolute error (MAE) 12.71 for the nonexpert group; ME 0.43 and MAE 5.57 for the expert group]. The intraclass correlation coefficient (ICC) for F-VASI was 0.45 [95% confidence interval (CI) 0.29-0.62] and 0.52 (95% CI 0.37-0.68) for nonexperts and experts, respectively. When the results were analysed separately for homogeneous and heterogeneous depigmentation, the ICC for homogeneous depigmentation was 0.47 (95% CI 0.31-0.77) and 0.85 (95% CI 0.72-0.94) for nonexperts and experts, respectively. When grading heterogeneous depigmentation, the ICC was 0.19 (95% CI 0.05-0.43) and 0.38 (95% CI 0.20-0.62) for nonexperts and experts, respectively.

CONCLUSIONS

We demonstrated that the Vitil-IA algorithm provides a reliable assessment of facial involvement in vitiligo. The study underlines the limitations of the F-VASI score when performed by nonexperts for homogeneous vitiligo depigmentation, and in all raters when depigmentation is heterogeneous.