We aim to evaluate the diagnostic performance of deep learning (DL) for determination of IDH mutation status in gliomas.

A systematic search of Cochrane Library, Web of Science, Medline, and Embase was conducted to identify relevant publications until 01/08/2021. Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) was used to assess studies' quality. Bayes theorem was utilized to calculate the posttest probability using likelihood ratios and predetermined pretest probabilities.

A total of four studies that included 1295 patients with glioma were included and analyzed in this study. The sample sizes for the training and validation sets were 1007 and 437, respectively. For training sets, using a known pretest probability of 80.2%, Bayes theorem yielded a posttest probability of 97.6% for a positive test and 27% for a negative test. On the other hand, a posttest probability of 96% for a positive test and 30.6% for a negative test were found for validation sets.

DL algorithms demonstrate an excellent diagnostic performance in predicting IDH mutation in gliomas. However, more studies are required to optimize and increase its reliability.