JMIR Publications

ABSTRACT

Background:

The detection and localization of cerebral microbleeds (CMBs) are significant for the diagnosis and treatment of the disease. Traditionally, the number and location of CMBs are manually calculated based on magnetic resonance imaging (MRI) characteristics such as shape, size, and signal features. While this method is highly accurate, experts must read the images and have significant prior knowledge, resulting in high detection costs. Therefore, it is necessary to explore an effective auxiliary detection method. In recent years, deep learning (DL) has received considerable attention in the detection of cerebral hemorrhage. Reviews have described the potential value of DL in the detection of CMBs. Some studies have explore image-based DL models for diagnosing CMBs. Nevertheless, systematic evidence regarding their diagnostic accuracy is lacking.

Objective:

This research intended to ascertain the accuracy of DL models in detecting CMBs, thereby offering evidence-based insights to facilitate the development of intelligent detection tools.

Methods:

This study was carried out following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and was prospectively registered in the International Platform of Registered Systematic Reviews (Registration ID: CRD42024628447). IEEE, Web of Science, Embase, the Cochrane Library, and PubMed were comprehensively searched up to November 1, 2024. The risk of bias of eligible studies was appraised utilizing the Quality Assessment of Diagnostic Accuracy Studies-2 tool. Subgroup analyses were implemented by image types and the generation method of the validation set.

Results:

In total, 29 studies were included, all of which developed models based on MRI. The meta-analysis results disclosed that the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were 0.94 (95% CI: 0.92–0.96), 0.94 (95% CI: 0.89–0.97), 15.3 (95% CI: 8.3–28.3), 0.06 (95% CI: 0.04–0.08), and 260 (95% CI: 109–624), respectively. In the random sampling validation set, the sensitivity, specificity, PLR, NLR, and DOR were 0.96 (95%CI: 0.92–0.98), 0.95 (95%CI: 0.89–0.98), 20.2 (95%CI: 8.6–47.6), 0.05 (95%CI: 0.03–0.08), and 438 (95CI: 115–1671), respectively. In cross-validation, the sensitivity, specificity, PLR, NLR, and DOR were 0.93 (95%CI: 0.91–0.96), 0.83 (5%CI: 0.61–0.94), 5.5 (95%CI: 2.2–14.2), 0.08 (95%CI: 0.05–0.12), and 70 (95%CI: 20–242), respectively. In the external validation set, the sensitivity, specificity, PLR, NLR, and DOR were 0.91 (95%CI: 0.86–0.95), 0.92 (95%CI: 0.81–0.97), 11.6 (95%CI: 4.5–29.9), 0.09 (95%CI: 0.06–0.15), and 123 (95%CI: 40–377), respectively.

Conclusions:

DL models based on MRI appear to show favorable diagnostic performance in detecting CMBs. Given the small number of included studies, more multicenter studies are warranted to facilitate the development of more generalizable detection tools. Clinical Trial: Registration ID: CRD42024628447