Abstract
BACKGROUND: We aimed to develop an artificial intelligence algorithm able to assess Raynaud's phenomenon (RP) from mobile phone photography, ensuring as a patient-centered, image-based method for RP quantification. METHODS: ARTIX (artificial intelligence-based Raynaud's quantification index) score was developed as a multi-step process of segmentation, decomposition and filters application to mobile phone pictures of the hand. ARTIX was validated by the ability to assess finger response to standardised cold challenge in patients with primary and secondary RP and healthy controls (HC) and compared with thermography as a reference. RESULTS: Forty-five RP patients (91.1% female, mean age 52.2 years, 75.5% secondary RP) were enrolled, along with 22 HC comparable for age and gender. RP patients presented significantly lower ARTIX values than HC both at baseline (p < 0.001) and across all timepoints of the cold challenge (p < 0.01 for all), paralleling a similarly significant difference observed by thermography. ARTIX score was higher in males and in patients taking vasoactive drugs, whereas lower values were obtained in patients with late capillaroscopic pattern, diffuse cutaneous skin subset, or negative for anti-centromere antibodies. ARTIX showed also good ability to discriminate between RP and HC response to cold challenge. CONCLUSION: We developed and validated ARTIX, a novel machine learning-driven method for the objective quantification of RP. Real-life longitudinal studies in patients with RP will determine the value of ARTIX to complement patient self-assessment surrogate measures of RP activity and severity.