P009 ALADIN: Automatic Lesion And Density INdex. A novel tool for automatic acne severity assessment

Abstract Acne is a common chronic disorder that impacts 85% of 16–18 year olds and 9.4% of the global population, regardless of age, gender or ethnicity. As of today, more than 25 scoring systems have been developed, but none has proven practical enough to generate a consensus among clinicians (Doshi A, Zaheer A, Stiller MJ. A comparison of current acne grading systems and proposal of a novel system. Int J Dermatol 1997; 36: 416–18; Bae IH, Kwak JH, Na CH et al. A comprehensive review of the Acne Grading Scale in 2023. Ann Dermatol 2024; 36: 65). This lack of methodological consistency contributes to observer variability and complicates acne monitoring and treatment evaluation. Lesion counting is the most direct and reliable method for severity assessment. While it is time-consuming, recent advancements in computer vision allow its automation with outstanding performance. This work leverages modern artificial intelligence to automate the task of acne severity assessment while providing explainable results. This study was retrospective in design. We trained a neural network to localize acne lesions in smartphone images. Based on the detected lesions, we analysed the density of the acne in the affected areas, which is strictly related to the severity of acne. Finally, we designed a new clinical endpoint, ALADIN (Acne Lesion And Density INdex), which integrates both lesion counting and lesion density to effectively assess acne severity while providing explainable results. We evaluated ALADIN against Investigator’s Global Assessment (IGA) and Global Acne Grading System severity ratings assessed by expert dermatologists. This validation was carried out on two different datasets: one with predominantly severe cases of acne from public atlases, and another one with predominantly mild-to-moderate cases from a private clinic. Notably, the ALADIN severity assessment achieved an IGA performance on par with expert dermatologists, with a Spearman correlation of 0.55 vs. 0.57, and a mean absolute error of 0.72 vs. 0.64. The final results confirm that ALADIN achieves a performance comparable with that of expert dermatologists, establishing it as a reliable and efficient tool for acne severity assessment. ALADIN is different from complex or subjective methods, and provides a simple and automated approach with the potential to standardize acne severity assessment and minimize interobserver variability.