Testing a signal-processing image analysis technology in hospital on suspicious pigmented naevi referred for biopsy

Four versions of the novel Moletest on-device signal-processing analysis software, delivered on a single-application iPadPro11, were evaluated in pigmented skin lesions of patients sent for biopsy comparing its results with the lesion histology. The primary objective was to demonstrate that the nomela^® test provided at least 95 ± 2% sensitivity for not-melanoma. Proprietary software recognised the lesion perimeter, rejecting images with uncertain border, and performed analysis using non-scalar metrics of the image within the perimeter. From five algorithms any one reading outside the predefined melanoma range assigned the lesion to 'no evidence of melanoma' or otherwise to 'melanoma not excluded'. 1365 participants provided 1573 lesions. 1389 valid cases had a valid signal-processing test result and a valid histology report defined as 'not-melanoma' (n = 1205) or 'melanoma' (n = 184). Statistical analysis for the four versions were: sensitivity (ability of the signal-processing analysis to assign 'No evidence of melanoma' to a lesion with 'not-melanoma' histology) 8.6/48.4/76.9/52.1%; specificity (ability to assign 'Melanoma not excluded' to a lesion with 'melanoma' histology) 100/79.6/75.0/65.2%. The performance of the Moletest signal-processing analysis as a diagnostic support for rule-out of melanoma showed limited value in all four versions.