JMIR Publications

ABSTRACT

Background:

In recent years, deep learning algorithms based on dermoscopy and whole slide imaging (WSI) have shown great potential in diagnosing basal cell carcinoma (BCC). However, the diagnostic performance of deep learning algorithms remains controversial.

Objective:

This meta-analysis evaluates the diagnostic performance of deep learning algorithms based on dermoscopy and WSI in detecting BCC.

Methods:

An extensive search through PubMed, Embase, and Web of Science was performed to locate pertinent studies published until November 4, 2024. A bivariate random-effects model was used to calculate the pooled sensitivity and specificity, both with 95% confidence intervals (CI). The heterogeneity of the studies was assessed using I2 statistic.

Results:

Among the 1,941 studies identified, 25 studies were included (internal validation sets of 56,873 patients or images, external validation sets of 963 patients or images). For dermoscopy-based deep learning algorithms, the pooled sensitivity, specificity, and the area under the curve (AUC) were 0.96 (95% CI: 0.93-0.98), 0.98 (95% CI: 0.96-0.99), and 0.99 (95% CI: 0.98-1.00). For dermatologists, the sensitivity, specificity, and AUC were 0.75 (95% CI: 0.66-0.82), 0.97 (95% CI: 0.95-0.98), and 0.96 (95% CI: 0.94-0.98). For WSI-based deep learning algorithms, the pooled sensitivity, specificity, and AUC were 0.95 (95% CI: 0.91-0.97), 0.97 (95% CI: 0.95-0.98), and 0.99 (95% CI: 0.98-1.00). For pathologists, the sensitivity was 0.88 (95% CI: 0.77-0.94), and the specificity was 0.82 (95% CI: 0.73-0.88).

Conclusions:

The results of this meta-analysis indicate that deep learning algorithms, whether based on dermoscopy or WSI, exhibit excellent diagnostic performance for BCC. Additionally, our findings show that dermoscopy-based deep learning algorithms outperform dermatologists in detecting BCC. These results highlight the potential of deep learning algorithms to assist human diagnosis of BCC in the future.