Cross-Modal Imaging in Noninvasive Identification of Histologic Features of Skin.

Abstract

Histopathology with light microscopy is the reference standard for cellular evaluation of solid tissue, but whether noninvasive cross-modal imaging of skin provides clinicians with comparable histologic information that may assist in clinical decision-making is not well known. To evaluate safety and effectiveness of cross-modal imaging in obtaining in vivo images, to demonstrate that images align with corresponding skin histopathology, and to evaluate the ability of blinded readers to accurately identify tissue features on cross-modal images. This observational diagnostic study included 1 visit and follow-up call. Eligible participants were adults aged 18 to 99 years seen for routine care in the outpatient dermatology setting with an indication for lesional skin biopsy. The study was conducted from October 20, 2022, to August 11, 2023, at 2 outpatient dermatology clinics in the US. Cross-modal images and skin biopsy were collected from 1 lesion per participant. Participants were randomized to training (40%) and validation sets (60%). Comparative readers used the training set to evaluate cross-modal images against hematoxylin and eosin histopathology to validate tissue features and train blinded physician readers. Comparative readers developed a performance test with the validation set; blinded physician readers were tested on cross-modal feature identification without access to other participant data. All participants underwent cross-modal imaging before biopsy. The primary end points were 100% comparative reader agreement and validation of cross-modal image tissue features against histopathology and 90% or greater accuracy of blinded physician readers identification of primary tissue features against comparative reader annotation. Secondary end points included accuracy of blinded physician reader identification of secondary features and greater than 90% interrater agreement between blinded physician readers. The safety end point was the number of adverse events observed. A total of 65 participants (median age, 69 [range 20-93] years; 41.5% female and 58.5% male; 1.5% American Indian or Alaska Native; 13.8% Hispanic or Latino; 86.2% not Hispanic or Latino; 98.5% White) underwent cross-modal imaging. Comparative readers achieved 100% consensus and validation of cross-modal histologic features compared with histopathology. Blinded physician reader accuracy for primary histologic features was 96.4% (95% CI, 94.2%-98.7%) and for secondary histologic features was 98.5% (95% CI, 98.1%-98.9%). Interrater agreement among the blinded physician readers was high (Fleiss κ: region, 0.94 [95% CI, 0.87-1.0]; feature, 0.93 [95% CI, 0.88-0.97]). No adverse events were reported. The results of this study suggest the safety of cross-modal imaging and that trained physicians may accurately identify histologic features on cross-modal images, consistent with its US Food and Drug Administration-cleared role in assisting clinical judgment.