Table 2.
Characteristics of studies that reported diagnostic accuracy of smartphone apps verified by histology with or without follow-up
| Study, country | Apps | No of patients, lesions | Inclusion criteria | Exclusion criteria | Data collection | Choice of lesions, image acquisition | Reference standard, target condition | No of exclusions (%) | No of cancers/total analysed, % (final diagnoses) |
|---|---|---|---|---|---|---|---|---|---|
| Chadwick (2014),23 Australia | SkinScan, Mel App, Mole Detective, Spot Mole Plus, Dr Mole Premium | NR, 15 | Images of melanocytic lesions excised with histopathological diagnosis | “Unable to analyse” lesions after 10 attempts | Retrospective (prospective interpretation) | Clinician, clinician | Histopathology (no FU), melanoma v benign naevus | Unevaluable images excluded a priori NR | 5/15, 33.3% (5 MM or MiS, 10 BN) |
| Dorairaj (2017),31 Ireland | App (not named*) | 32, 32 | Patients referred for excision of pigmented lesions | NR | Prospective | NR, clinician | Histopathology (no FU), melanoma v dysplastic naevus or benign | Unevaluable images 6 (19%) | 9/26, 35% (9 MM or MiS, benign diagnoses NR) |
| Maier (2015),28 Germany | SkinVision (original version) | NR, 195‡ | Patients with melanocytic skin lesions seen routinely for skin cancer screening at dermatology department | Poor quality index test image; other elements in the image, eg, hair, images containing more than one lesion, incomplete imaged lesions, non-melanocytic lesions, two point differences cases, tie cases | Prospective | Clinician, clinician | Histopathology (no FU), melanoma v not melanoma | Unevaluable images 20 (10%); “tie” cases 18 (9%); two point differences 13 (7%) | 26/144, 18.1% (26 MM or MiS, 34 DN, 84 BN) |
| Robson (2012),32
United Kingdom |
MelApp | 31, 35 | Patients with pigmented skin lesions referred from GPs to urgent cancer clinic | NR | Prospective | NR, clinician | Histopathology (49%) or clinical assessment (no FU), melanoma v not melanoma | Unevaluable images 14 (40%) | 2/21, 9.5% (histology: 2 MM, 4 BN, 2 DN, 1 blue naevus, 2 SK; clinically benign 10) |
| Thissen (2017),33 Netherlands | SkinVision (original and rev±qu)† | 256, 341§ | Patients with pigmented or non-pigmented skin lesions seen routinely at the dermatology department | NR | Prospective | Clinician, clinician | Histopathology (38%) or clinical assessment (no FU), malignant or premalignant v benign | None excluded owing to quality to mimic real world use | 35/108, 32.4% (malignant or premalignant: 2 MM, 1 MiS, 16 BCC, 3 cSCC, 5 BD, 8 AK, all with histology apart from 2 BCC and 3 AK; benign: 9 SK, 12 BN, 1 DN, 7 SL, 3 LPLK, 41 other benign¶) |
| Wolf (2013),34
United States |
3 Apps (not named) | NR, 188 | Images of pigmented skin lesions with a clear histological diagnosis assessed by a board certified dermatopathologist | “Difficult to diagnose” lesions, lesions with equivocal diagnoses, specific lesion types, eg, SN or atypical naevi (moderate high grade), images with identifiable features | Retrospective (prospective interpretation) | Clinician, clinician | Histopathology (no FU), melanoma v benign lesions | Unevaluable images app 1: 6 (3%); app 2: 3 (2%); app 3: 17 (9%); plus poor quality excluded a priori | 60/188, 31.9% (44 MM, 16 MiS, 94 BN, 20 SK, 8 SL, 2 hemangiona 2, 4 DF) |
The Chadwick 2014 SkinScan app is a predecessor of SkinVision and not related to the CE marked TeleSkin skinScan app. AK=actinic keratosis; BCC=basal cell carcinoma; BD=Bowens disease; BN=benign naevi; cSCC=cutaneous squamous cell carcinoma; DF=dermatofibroma; DN=dysplastic naevi; FU=follow-up; GP=general practitioner; LPLK=lichen planus like keratosis; MiS=melanoma in situ; MM=malignant “invasive” melanoma; NR=not reported; SK=seborrhoeic keratosis; SL=solar lentigo; SN=Spitz naevi.
App name not declared in study and could not be divulged by the authors (J Dorairaj, personal communication, 2019).
SkinVision (rev±qu)=revised version of the SkinVision app, “recalibrated” to accommodate non-pigmented lesions and including the option of a questionnaire about lesion characteristics (eg, texture, colour, shape, size, and symptoms).
At least three images per lesion.
108/341 included as test set.
41 other benign include 10 psoriasis, 8 histiocytoma, 8 folliculitis, 3 sebaceous hyperplasia, 6 angioma senilis, 4 scars, 1 clear cell acanthoma, 1 verruca vulgaris.