Logo image
Back
Monitoring of Melanocytic Skin Lesions Incorporating Artificial Intelligence to Detect Melanoma in Outdoor Enthusiasts
Thesis   Open access

Monitoring of Melanocytic Skin Lesions Incorporating Artificial Intelligence to Detect Melanoma in Outdoor Enthusiasts

Ian Miller
Southern Cross University
Doctor of Philosophy (PhD), Southern Cross University
2025
DOI:
https://doi.org/10.25918/thesis.559
Appears in  Recent Southern Cross PhD Theses
pdf
Miller I PhD 202510.45 MBDownloadView
Open Access

Metrics

6 Record Views

Abstract

Physical activity Melanoma Ultraviolet radiation Skin cancer Artificial Intelligence Keratinocyte cancer Skin Neoplasm Sports
The objectives of this research were to assess the prevalence of skin cancer in surfers, swimmers, walkers/runners and investigate performance of artificial intelligence (AI) for classifying lesions of interest suspect of being melanoma; systematically analyze the literature reporting on the performance metrics of market-approved AI in melanoma classification; investigate the performance of a market-approved convolutional neural network (CNN) to better differentiate skin lesions suspicious of being melanoma; and investigate the prevalence of aquatic and non-aquatic participants and the locations of these skin cancers. For objective one, a cross-sectional study design incorporating a survey, total-body skin cancer screen and AI-embedded software capable of reporting a predictive risk score for queried melanoma was incorporated. For objective two, a systematic literature search was performed. For objective three, a multi-center, cross-sectional study using a commercially available CNN on lesions of interest from participants attending a skin examination within two Australian clinics. The fourth and summarising objective was met by a combination of participant survey and total-body skin screen to determine prevalence for keratinocyte cancer and melanoma. Rates of both keratinocyte carcinomas and melanoma were notably higher in surfers, swimmers, walkers and runners compared to the general Australian population. Lower-than-expected performance metrics of AI software in an Australian skin cancer focused clinic indicate further investigation of this software on a larger dataset is warranted. The systematic review of the literature yielded 16 full-text articles. Performance metrics for classifying melanoma were highly heterogeneous with sensitivity, specificity and accuracy ranging between 16.4% to 100%. Two different general practice clinics in Australia with a focus on skin cancer detection reported comparable performance metrics. Of note, accuracy was calculated as 56.6% (Townsville 56.1%, Gold Coast 57.5%) which is notably lower than comparable studies in American and European-based locations. Melanoma was found to have an odds ratio 1.8 times higher for aquatic participants compared to non-aquatic participants. Participants of increased age and male gender had an increased likelihood of being diagnosed with melanoma. Primary locations for keratinocyte cancer were the head, face and neck which differed from melanoma, being primarily located on the back. This thesis provides real-world evidence on how AI performs when running as an adjunct within general practice clinics tasked with early detection of skin cancers. Outdoor enthusiasts should be recognised as a cohort with a high risk for melanoma which should be reflected in a national screening policy, and preventative actions advised by clinicians.

Details

Logo image