MoleGazer Development Feasibility Study

Melanoma (Skin) Clinical Trial

Official title: MoleGazer: A Feasibility Study for Early Detection of Melanoma

Status Recruiting

Clinical Trial Summary

Melanoma (skin cancer) frequently develops from existing moles on the skin. Current practice relies on expert dermatologists being able to successfully identify new/changing moles in individuals with multiple moles. Total body photography (TBP-high-quality images of the entire skin) can track and monitor moles over time to detect melanoma. However, TBP is currently used as a visual guide when diagnosing melanoma, requiring visual inspection of each mole sequentially. This process is challenging, time-consuming and inefficient. Artificial intelligence (AI) is ideally suited to automate this process. Comparing baseline TBP images to newly acquired photographs, AI techniques can be used to accurately identify and highlight changing moles, and potentially distinguish harmless moles from cancerous changes. Astrophysicists face a similar problem when they map the night sky to detect new events, such as exploding stars. Using AI, based on two or more images, astrophysicists detect new events and accurately predict how they will appear subsequently. This project, called MoleGazer, is a collaboration with astrophysicists aiming to apply AI methods that are currently used for astronomical sky surveys, to TBP images. The MoleGazer algorithm, developed at Oxford University Hospitals NHS Foundation Trust, will automatically identify the appearance of new moles and characterise changes in existing ones, when new TBP images are taken. To optimise this MoleGazer algorithm TBP images will be taken at multiple time-points, as there are no existing datasets of TBP images that are publicly available. The investigators invite a) high-risk patients attending skin cancer screening clinics to attend sequential three-monthly TBP imaging and clinical assessment and b) any patient who undergoes TBP as standard care to share images so that the investigators can develop the MoleGazer algorithm. The ultimate goal is for the MoleGazer algorithm to 'map moles' over a patient's lifetime to detect changes, with the eventual aim to detect melanoma as early as possible.

Clinical Trial Description

Background Melanoma incidence is rapidly increasing with 15,906 new United Kingdom (UK) cases in 2015 resulting in 2,285 deaths. Diagnosing melanoma early is essential as early stage disease has > 95% 5-year relative survival rate compared with 8-25% for advanced melanoma. In the UK, skin cancer costs are predicted to exceed £180 million by 2020 and pose significant morbidity (and mortality) to individuals affected. Up to 60% of melanoma arise from pre-existing naevi (moles). Early melanoma detection relies on individuals recognising changes in naevi and for those individuals with multiple naevi expert assessment of these naevi by trained dermatologists using diagnostic aids such as dermoscopy (x10 magnification). Furthermore there is evidence that sequential surveillance of naevi also increases melanoma detection rates. Total body photography (TBP) is a diagnostic aid for monitoring of multiple naevi For patients at high-risk of developing melanoma with multiple naevi (>60), total body photography (TBP) (standardised body-part images taken using high-resolution camera), is used as an aid to track, compare and monitor naevi over time and has been demonstrated to improve melanoma diagnosis. Recommended short-term surveillance monitoring of naevi is 3-months but is largely confined to single lesions. In a resource-constrained National Health Service (NHS), frequent surveillance for multiple naevi by a dermatologist is impractical and inefficient such that early diagnosis of melanoma effectively relies on patient self-surveillance. A potential solution is automated analysis of TBP images using artificial intelligence (AI) to track and monitor naevi over time. Artificial intelligence applied to TBP could improve efficiency of 'mole-mapping' Previous AI evaluation of skin lesions has demonstrated equivalent accuracy to trained dermatologists in skin cancer diagnosis, however this relied on single-lesion analysis at static time-points (with biopsy-proven diagnoses). The use of lesions scheduled for excision (i.e., high clinical suspicion of melanoma) severely limits clinical applicability and a Cochrane review concluded that utility of computer-aided detection for melanoma diagnosis in secondary care remains unknown.The more clinically-relevant question is whether automated detection of changes in naevi using sequential TBP images, referred to clinically as 'mole mapping', can indeed improve early diagnosis of melanoma. To date, TBP systems in the NHS have limited automation, restricted to storing and retrieving images. Although one automated total body scanning system exists, and in the future may incorporate AI-based diagnosis in addition to current image acquisition and lesion matching algorithms, a full clinical validation and any subsequent implementation in the NHS will be costly due to the investment required in the scanning system (current cost US $1 million). Whether the same or better results can be achieved using more conventional image acquisition equipment and sophisticated AI techniques is unknown. The investigators propose a novel application of astronomical AI methods for early melanoma detection using standard TBP-based surveillance of naevi which is currently employed in the NHS and can be used as an adjunct to clinical review of individuals. Application of astronomical AI techniques to TBP monitoring of multiple naevi Transient science in astronomy aims to detect and track evolution of new astronomical sources such as exploding stars. Exhibiting both long- and short-term evolution, individual events are detected by comparing new images with archival data and classified based on a feature set, including transient brightness, colour, proper motion and extent. Cutting-edge astronomical surveys monitor the sky every night over multi-year timescales to identify subtle changes. AI techniques (such as random forests and recurrent neural networks; RNN) which use the full time-series history and contextual information are routinely used to identify and classify events probabilistically. With each new observation providing additional information, astronomical transient surveys can routinely detect and characterise new sources, such that the evolution of new sources can be predicted with 99.5% accuracy based on only three time-points. This challenge faced in astronomy is analogous to 'mole mapping' for individuals at high-risk of developing melanoma; both naevi and astronomical sources can be characterised as distinct sources against a homogeneous background which are tracked across multiple images to detect change. The investigators therefore hypothesise that astronomical AI techniques are ideally suited to address this clinical problem and are developing the MoleGazer project to test this. Rationale To develop the MoleGazer algorithm, the investigators require a baseline dataset to apply astronomical AI algorithms to TBP images to detect and track naevi across sequential images. There are currently no publicly available databases of TBP images for the investigators to test this feasibility and therefore in this study the aim is to collect: 1. a time-series cohort of TBP images taken at fixed sequential time-points over 2 years 2. a baseline cohort of TBP images with sequential images taken at any time-points By collecting TBP images it will allow the investigators to study the sensitivity of naevi detection and characterisation on skin tone, lighting levels, image registration and background subtraction techniques, enabling the investigators to also automate detection of naevi and track their evolution in any sequential image that the study team has. The development of this database will allow the investigators to demonstrate feasibility of the application of astronomical AI methods to TBP images. ;

Related Conditions & MeSH terms

• Melanoma
• Melanoma (Skin)

• NCT number: NCT05015816
• Study type: Interventional
• Source: Oxford University Hospitals NHS Trust
• Contact: Rubeta N Matin, PhD FRCP
• Phone: +441865 228264
• Email: rubeta.matin@ouh.nhs.uk
• Status: Recruiting
• Phase: N/A
• Start date: September 13, 2021
• Completion date: February 12, 2026