Performance Evaluation of Artificial Intelligence Techniques in the Diagnosis of Brain Tumors: A Systematic Review...

Brain tumors are becoming more prevalent, often leading to severe disability and high mortality rates due to their poor prognoses. Early detection is critical for improving patient outcomes. These tumors pose substantial diagnostic challenges because of their varied manifestations, necessitating timely and accurate diagnosis. Recent advancements in artificial intelligence (AI) have shown the potential to enhance diagnostic accuracy, especially through MRI analysis. We analyzed the performance of AI algorithms for various types of tumors as well as for different diagnostic goals, with special consideration of assessing the accuracy, precision, recall, and F1 score of AI for recognizing gliomas, meningiomas, and pituitary tumors, as well as for identifying tumor versus non-tumor tissue. By integrating both the performance metrics and the methodology used, this review offers an overall comparative analysis of AI-based diagnostic methods on brain tumor images. This study aims to systematically review the use of AI techniques in diagnosing brain tumors through MRI scans, following the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines. We conducted a search across multiple databases, including PubMed, Embase, Web of Science, Cochrane Library, and Scopus. Our search encompassed publications from 2000 to February 2024. In total, we identified 79 studies that met our inclusion criteria. These criteria required the use of MRI for brain tumor detection and classification, and the utilization of clearly defined performance metrics such as precision, recall, F1 score, accuracy, sensitivity, and specificity. To assess the quality of the studies, we employed the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool. Our meta-analysis specifically focused on evaluating the performance of different algorithms in relation to various types of brain tumors. The analysis incorporated data from seven selected articles. The meta-analysis shows that AI methods accurately diagnose brain tumors using MRI. The overall F1 score ranges from 0.945 to 0.958, with an estimated accuracy of 0.952. The top performers in this field are convolutional neural networks (CNNs), ensemble algorithms, and support vector machines. Among these, CNNs have a slightly higher F1 score (0.953) compared to ensemble algorithms (0.949). The accuracy varies depending on the tumor type, with gliomas having an F1 score of 0.961, pituitary tumors at 0.955, and meningiomas at 0.950. The meta-regression analysis reveals that tumor type significantly influences accuracy, with lower scores observed in the "tumor/no tumor" category. AI models demonstrate high diagnostic accuracy in controlled research settings (pooled accuracy: 0.952, 95% CI: 0.945-0.958), but significant heterogeneity (I²=40.75%) and performance variability across tumor types limit broad clinical generalizability. CNNs and ensemble algorithms show consistent results for glioma and pituitary tumors, but evidence for clinical deployment remains preliminary.