ACCURACY OF AI-BASED SKIN CANCER DETECTION APPS IN IDENTIFYING MELANOMA

Abstract

The significance of early recognition of melanoma lies in enhancing patient survival, and decreasing the later disease burden due to the worsened prognosis once the incidence progresses. Diagnosis mostly depends on clinic examination taking place with dermatologists and histopathological confirmation that can take time, money and may not be available in some regions. With improvement of artificial intelligence (AI) and machine learning now people have a chance to get mobile applications which are able to analyze dermoscopic and clinical images for skin malignancy risk assessment. AI-powered instruments make it probable to carry out fast, non-invasive preliminary screening; increase patients’ access to dermatology screening, as well as aid healthcare providers in making clinical decisions. This present study was aimed at determining how effective three popular AI-based skin cancer detection programs are by evaluating their sensitivity, specificity, and overall accuracy in diagnosing melanoma. A cross-sectional study utilizing 500 de-identified dermoscopic images was carried out; 250 of them were histopathologically confirmed melanomas while the other 250 were benign lesions including nevi and seborrheic keratoses. Every image was submitted for analysis by all three AI applications, then its results were juxtaposed with histopathological diagnosis’s gold standard. The applications had sensitivity levels oscillating between 82% up to 91%, specificity levels fluctuating between 76% till reaching 88%, total accuracy ranging from 79% till arriving at 86%.