Abstract

Background: Recently, a growing body of evidence has been amassed on evaluation of artificial intelligence (AI) in computer-aided diagnosis of cutaneous lesions by means of deep learning algorithms. Objective: We conducted this meta-analysis to study the pooled rates of performance for AI in diagnosis of cutaneous lesions based on histopathological whole slide image analysis. Methods: Multiple databases were searched (from inception to May 2021) and studies that reported on the performance of AI in the diagnosis of cutaneous lesions on histopathology were selected. A random effects model was used, and pooled accuracy were calculated. Results: Four studies evaluating six deep learning algorithms were included in our final analysis. Olsen TG et al evaluated the utility of three deep learning algorithms in the diagnosis of dermal nevi, basal cell cancer and seborrheic keratosis. Hart SN et al applied a convolutional neural network (CNN) to differentiate between two forms of melanocytic lesions (Spitz and conventional), Ianni JD et al looked at an artificial intelligence algorithm for melanoma recognition and Brinker TJ et al evaluated a CNN algorithm to classify skin cancers on histopathology. The pooled accuracy of the six deep learning algorithms based on random effects model was 97.3 % (95% confidence interval = 93.4-98.9) with a heterogeneity (I-squared) of 89.2. The publication bias was acceptable based on funnel plot analysis of standard errors. The sensitivity ranged from 85-94% and specificity was 90-99%. Conclusions: Based on our meta-analysis, AI achieved high accuracy in diagnosis of lesions in skin and can be a potential aid in diagnosis for dermatopathologists.

Financial Disclosure: No current or relevant financial relationships exist.