Accuracy of tele‐consultation on management decisions of lesions suspect for melanoma using reflectance confocal microscopy as a stand‐alone diagnostic tool

Abstract

Background Diagnostic accuracy of reflectance confocal microscopy (RCM) as a stand‐alone diagnostic tool for suspect skin lesions has not been extensively studied. Objective Primary aim was to measure experts’ accuracy in RCM‐based management decisions. Secondary aim was to identify melanoma‐specific RCM features. Methods The study enrolled patients ≥18 years that underwent biopsy of skin lesions clinically suspected to be melanoma. One hundred lesions imaged by RCM were randomly selected from 439 lesions prospectively collected at four pigmented lesion clinics. The study data set included 23 melanomas, three basal cell and two squamous cell carcinomas, 11 indeterminate melanocytic lesions and 61 benign lesions including 50 nevi. Three expert RCM evaluators were blinded to clinical or dermoscopic images, and to the final histopathological diagnosis. Evaluators independently issued a binary RCM‐based management decision, ‘biopsy’ vs. ‘observation’; these decisions were scored against histopathological diagnosis, with ‘biopsy’ as the correct management decision for malignant and indeterminate lesions. A subset analysis of 23 melanomas and 50 nevi with unequivocal histopathological diagnosis was performed to identify melanoma‐specific RCM features. Results Sensitivity, specificity and diagnostic accuracy were 74%, 67% and 70% for reader 1, 46%, 84% and 69% for reader 2, and 72%, 46% and 56% for reader 3, respectively. The overall kappa for management decisions was 0.34. Readers had unanimous agreement on management for 50 of the 100 lesions. Non‐specific architecture, non‐visible papillae, streaming of nuclei, coarse collagen fibres and abnormal vasculature showed a significant association with melanoma in the evaluation of at least two readers. Conclusions Reflectance confocal microscopy tele‐consultation of especially challenging lesions, based on image review without benefit of clinical or dermoscopy images, may be associated with limited diagnostic accuracy and interobserver agreement. Architectural and stromal criteria may emerge as potentially useful and reproducible criteria for melanoma diagnosis.

Reflectance confocal microscopy diagnostic accuracy for malignant melanoma in different clinical settings: systematic review and meta‐analysis

“…Thirty‐one studies were excluded because: (i) MM RCM diagnostic accuracy data were missing/studies including only morphologic descriptions of single RCM features ( n = 9), (ii) MM other than cutaneous ( n = 8) and (iii) <10 lesions were evaluated ( n = 14). Finally, 34 studies (7512 lesions) were included in the systematic review . Two studies were excluded from meta‐analysis because data for the calculation of MM specificity and sensitivity were insufficient.…”

“…The systematic review included 34 observational studies, published between 2005 and 2018. Ten studies (29%) were performed prospectively (three interventional and seven non‐interventional) and 24 (71%) retrospectively (Table ). Of interest, 14 studies (41%) included other types of skin cancers, 11 (32%) were multicentre studies, and 6 (18%) reported RCM evaluations performed in a telemedicine setting .…”

“…Ten studies (29%) were performed prospectively (three interventional and seven non‐interventional) and 24 (71%) retrospectively (Table ). Of interest, 14 studies (41%) included other types of skin cancers, 11 (32%) were multicentre studies, and 6 (18%) reported RCM evaluations performed in a telemedicine setting . All studies included lesions without a straightforward clinical or dermoscopic diagnosis and required second‐level evaluation; of these, 20 (59%) studies focused on equivocal‐difficult lesions highly suspicious for MM, while 14 studies (41%) included lesions consecutively referred for evaluation in a clinical setting .…”

“…Seven studies (522 lesions) were prospective non‐interventional, reporting a sensitivity of 97% (95% CI: 0.94–1, I 2 = 31%) and specificity of 82% (95% CI: 0.77–0.86, I 2 = 81%). Twenty‐two retrospective studies (4,855 lesions) reported a sensitivity of 90% (95% CI: 0.88–0.92, I 2 = 65%) with a specificity of 73% (95% CI: 0.72–0.75, I 2 = 93%). The three prospective interventional studies (1975 lesions) reported a sensitivity of 96% (95% CI: 0.94–0.97, I 2 = 83%) with a specificity of 59% (95% CI: 0.57–0.62, I 2 = 98%).…”

Utilizing Machine Learning for Image Quality Assessment for Reflectance Confocal Microscopy

“…Therefore, real-time quantitative assessment of image quality has a key role in image capture. The image capture capabilities of clinical RCM devices, including mosaics and image stacks, have been described previously (Scope et al, 2019). Mosaics are captured at several layers in the epidermis and superficial dermis as part of a standard image set, followed by acquisition of image stacks.…”