Investigation of diagnostic value of artificial intelligence systems in the diagnosis of breast cancer based on histopathological images using meta-MUMS DTA tool

Abstract

Background: Various artificial intelligence systems are available for diagnosing breast cancer based on histopathological images. Assessing the performance of existing methodologies for breast cancer diagnosis is vital.
Methods: The SCOPUS database has been searched for studies up to December 15, 2018. We extracted the data, including "true positive," "true negative," "false positive," and "false negative". The pooled sensitivity, pooled specificity,
positive likelihood ratio, negative likelihood ratio, diagnostic odds ratio, area under the curve of summary receiver operating characteristic curve were useful in assessing the diagnostic accuracy. Egger's test, Deeks' funnel plot, SVE
(Smoothed Variance regression model based on Egger’s test), SVT (Smoothed Variance regression model based on Thompson’s method), and trim and fill methodologies were essential tests for publication bias identification.
Results: Three studies with eight approaches from thirty-seven articles were found eligible for further analysis. A sensitivity of 0.95, a specificity of 0.78, a PLR of 7525, an NLR of 0.06, a DOR of 88.15, and an AUC of 0.953
showed high significant heterogeneity; however, the reason was not the threshold effect. The publication bias was detected by SVE, SVT, and trim and fill analysis.
Conclusion: The artificial intelligent (AI) systems play a pivotal role in the diagnosis of breast cancer using histopathological cell images and are important decision-makers for pathologists. The analyses revealed that the
overall accuracy of AI systems is promising for breast cancer; however, the pooled specificity is lower than pooled sensitivity. Moreover, the approval of the results awaits conducting randomized clinical trials with sufficient data