ARTIficial Intelligence-based Smartphone Application for Skin Cancer Detection

Skin cancer is the most frequent cancer diagnosed and its incidence will keep on rising in the next decade. Early detection and treatment are key to improve both morbidity and mortality, and to decrease the cost to society. Persons at risk of developing skin cancer may be subjected to regular checkups. However a considerable number of skin cancers develop in the low-risk general population. Since systematic screening in the general population is not cost-effective, smartphone applications that use inbuilt algorithms are of increasing interest and claim to assist in making a risk assessment in case of concerning skin lesions.

Based on previous research, a so-called triage consultation was installed at the policlinic of Ghent University Hospital for patients with 1 to 2 lesions of concern: changing mole, ugly duckling, new mole in adult, rapidly growing lesion or non-healing lesion. Skin cancer detection rate in this setting was at least 13% with 4% melanoma. This is 6 to 8-fold higher than reported by conventional skin cancer screening programs (PMID: 26466155; PMID: 33480073). The reason for this is that a preselection of lesions meeting specific criteria is done. This lesion-directed screening may be a way to make skin cancer screening in the general population (more) cost-effective.

In this study we will investigate whether the Skinvision app can function as a preselection tool for lesions for which urgent medical advice is needed. Although this app is CE marked and is already promoted to the public, it's performance and value in daily practice have been insufficiently studied and there is a need for independent research.

The 4 main objectives of this study will be:

1. To calculate diagnostic performance of the Skinvision App Calculation of sensitivity and specificity by comparing application risk gradings with a reference standard defined as the histopathological diagnosis or clinical diagnosis in case no biopsy or excision was performed;

2. To determine the repeatability and reproducibility of the Skinvision App Identification of factors that influence the risk analysis of the application, including photographer, type of skin lesion, camera position or lighting conditions;

3. To examine user-experience and confidence concerning the use of medical apps Questionnaire-based evaluation of the user-experience with applications in general, as well as more specific the willingness and confidence to use a skin cancer detection application;

4. To estimate the performance and cost-effectiveness of the Skinvision App in the general population Estimation of the app performance in the general population (estimated prevalence of skin cancer 1%) in terms of missed diagnoses and degree of preselection (positive predictive value).