Moletest (Scotland) Ltd has over 10 years of experience in developing the technical, clinical and regulatory programmes of nomela®.

The initial analytical system (2013) used AI, trained from publicly available library images and a large pre-determined group of algorithms, with fuzzy logic to make the assessment for malignancy. However, when made available to the public on-line, the system was found to be unreliable because users took poor quality images.

The analysis engine was replaced (2014) by five selected proprietary signal-processing algorithms in nomela®, an image capture and analysis system for use by trained medical professionals, embedded on a single-application iPad. Although found to be easy to use by dermatology nurses in clinical studies the performance of the analytical engine was considered lower than expectation.

The analytical engine used in nomela® v6 is a machine-learning AI system, selected after substantial and rigorous comparative testing. Training and validation was made on images derived from a basic set of c1000, all with consent and definitive histopathology results obtained in a clinical study. The nomela® v6 AI system is uniquely enhanced by patented automated edge detection to ignore skin surrounding the lesion under test.

Summary of clinical studies
and technical aspects

Technical note

Proprietary signal processing was used as the analytical engine in the clinical studies of nomela® (004/C4[2015-2016]; C7 [2018]; C8 [2019-2022]) and was formally set aside in 2023 (see below).

Change of the analytical engine to a machine-learning AI has been in research since 2022. Participant recruitment in clinical study C8 was suspended in 2022. Return to prospective testing of nomela®v6, the AI engine, is planned.

nomela® 004/C4 [2015-2016]

A hospital-based study of a novel photographic image analysis system to discriminate benign pigmented skin lesions from those considered suspect for melanoma.

IRAS project ID: 159049; NHS Research Ethics Committee, Cambridge South 15/EE/0097; NHS Lanarkshire L14100; NCT02196246

The study aimed to assess performance in discriminating pigmented skin lesions as benign from among those considered suspicious of skin malignancy including melanoma in people referred from primary care for specialist dermatology review of suspicious pigmented moles.

A total of 1200 participants were recruited. Photographic images were taken by the Medical Illustration Service, using a dedicated iPad Air2 loaded with the nomela® technology.

Excluding 195 lesions for clinical, technical or data capture reasons, and using the 716 lesions with all 5 signal processing parameters available, nomela® was able to exclude melanoma in 61.2% (at least one measurement beyond the range for ‘melanoma’) (discriminator D1). Of these 438 lesions, 346 were benign [79.0%], 59 dysplasia [13.5%] and 33 [7.5%] other malignancies. To use the D1 discriminator would mean over half of lesions otherwise unassessed could exclude melanoma with four-in-five in this group being benign but with a one- in-five chance of dysplasia or non-melanoma malignancy.

nomela® was able to exclude melanoma and other skin malignancy (discriminator D2) in 21.9%/just over one-in-five of lesions, or, alternatively, (discriminator D3) in about 15.9%/one-in-six of lesions. Using D2, the low risk group contains c81% benign and c19% dysplasia. Using D3, the low risk group contains 88% benign and 12% dysplasia.

The clinical utility of nomela® on the basis of this evidence is that, with the cautions regarding relatively low absolute numbers of malignancies in this study, it should be possible to deploy the technology in trained professional hands to assist in the diagnosis of pigmented moles and may reduce the need for specialist review by 15-20% albeit with appropriate advice and caution.

nomela® C7 [2018]

Testing of nomela® on image of skin lesions confirmed as cutaneous melanoma.

IRAS project ID: 242101; NHS Research Ethics Committee, West of Scotland 5 18/WS/0065; NHS SSI 343880; NHS Lanarkshire L18002; ISRCTN63318109

The study aimed to identify and obtain consent for use of historic images of confirmed cutaneous melanoma which had been collected in the normal clinical pathway for subsequent retrospective examination with nomela®. There was no impact on clinical care of the persons giving consent.

Over the defined 24-month period the number of usable images-with-consent was 99 after technical exclusions, absence of images and non-response to requests for consent. The melanoma types were mainly superficial spreading melanoma (66) or melanoma-in-situ /superficial spreading melanoma-in situ (22).

Technical note

The images obtained in C7 were used together with those from 004/C4 to refine ranges for the 5 signal processing parameters for not-melanoma.

Technical reevaluation demonstrated that the test set ‘New Bounds’ (116 melanoma and 424 ‘not-melanoma’ lesion images) using an ‘auto-crop’ facility (a fundamental change in the way the edge of the lesion was followed for all images) provided the greatest improvement in performance, and that nomela® was capable of differentiating 53% images as ’no evidence of melanoma’.
The technical outcome was nomela®v2.

See BAD Poster 288 (2020)

nomela® C8 [2019 - ]

Testing nomela® on suspicious pigmented naevi: A hospital-based study.

IRAS project ID: 254451; NHS Research Ethics Committee, Cambridge Central 19/EE/0041; NIHR CRN DERM40826; Cambridge University Hospitals R&D A094925; ISRCTN99987356

The study was designed to measure the performance of the nomela® test on pigmented skin lesions of patients referred by specialists for biopsy/excision by comparing the test results with the subsequent histopathology results. The nomela® test results were not used for clinical management.

Target number for valid cases was 200 completed nomela® test results linked to a histology result of melanoma. It was predicted that c1600 participants each with a single lesion would need to be recruited and tested, at the point of decision to proceed to biopsy, based on the expected ratio of not-melanoma/melanoma histology results in the study population.

Five versions (V2, V3, V4.0, V4.1 and V5.0) of the nomela® analysis were deployed during Study. This has been used to define the five distinct Periods 1 to 5. Limits for melanoma in the nomela® test software were different in each of the five periods. Further, definition of the lesion edge was by operator manipulation in V2 and V3 and by proprietary automated edge detection in V4.0, V4.1 and V5.

Valid cases, defined as having both a usable test result "No evidence of melanoma" or "Melanoma not excluded” according to the contemporary version of the nomela® test analysis and usable histology report (‘not-melanoma’ or ‘melanoma’) in Periods 1 to 4 was 1275 [all periods: 1374]. Of these completed nomela® analyses the result was "No evidence of melanoma" in 487 [all periods: 501] and "Melanoma not excluded” in 788 [all periods: 873].

Conclusion: the nomela® test on the nomela® device satisfactorily captured 1580 images of pigmented skin lesions and provided a result for 85% of tests (Periods 1 to 4). Lesions calculated as too small (approximately less than 5 mm diameter) were 7.7% and having an indistinct edge were 7.3%.

The best performance of the signal processing software was demonstrated in Period 3 using version 4.0 with a sensitivity 76.9 % (95%CI 68.2 - 85.5) and specificity 75 % (95%CI 56.1 - 93.8) based on 111 valid cases of which 20 were melanoma.

The results of the signal-processing software analysis in the four versions deployed (V2, 3, 4.0 and 4.1) indicated limited evidence for the ability of the nomela® software in these versions to provide useful additional diagnostic support in the assessment of suspicious pigmented moles.

Note: Study C8 remains open for software development to allow for a restart of participant recruitment (agreed as “suspended” with NIHR CRNCC reset 2022.10.05).