High-definition optical coherence tomography algorithm for the discrimination of actinic keratosis from normal skin and from squamous cell carcinoma.

BACKGROUND

Preliminary studies described morphological features of actinic keratosis (AK) and squamous cell carcinoma (SCC) imaged by High-Definition Optical Coherence Tomography (HD-OCT) and suggested that this technique may aid in their diagnosis. However, systematic studies evaluating the accuracy of HD-OCT for the diagnosis of AK and SCC are lacking so far.

OBJECTIVE

In this study, we sought to design an algorithm for AK classification that could (i) distinguish SCC from AK and normal skin, (ii) differentiate AK from normal skin and (iii) discriminate AKs with adnexal involvement from those without.

METHODS

A total of 53 histopathologically confirmed lesions (37 AKs and 16 SCC) were imaged by HD-OCT. Fifty-three HD-OCT images of normal skin of healthy volunteers, with matched age, skin type and anatomic site, were taken as reference. By comparing these 106 en face and cross-sectional HD-OCT images, particular features were selected based on their potential to discriminate AK from normal skin and from SCC, and to assess adnexal involvement in AK. This study represents a training set not a testing set. Severe (>300 μm) hyperkeratotic AKs were not included in this study.

RESULTS

Particular features with high Phi coefficient could be identified. The absence of an outlined dermo-epidermal junction (DEJ) on cross-sectional images allowed discriminating SCC from AK and normal skin (Phi coefficient = 0.84). AK could be discriminated from normal skin in both imaging modes by the presence of alternating hyperkeratosis/parakeratosis in cross-sectional mode and/or variability in shape, size and reflectivity of cells (atypical honeycomb pattern) in en face mode. Adnexal involvement of AK could be assessed by the disappearance of the typical cocarde image of adnexal epithelium in en face mode.

CONCLUSION

This study provides select 3-D HD-OCT features having a potential to discriminate SCC from AK and normal skin. Based on these particular features with high Phi coefficient, a diagnostic algorithm is designed which will be used later in validation studies to determine HD-OCT accuracy in AK/SCC classification.