Meta-analysis of Artificial Intelligence for pCLE-based brain tumour characterisation and extent of tumour resection

<h3>Introduction</h3> Addressing glioblastoma, which remains the most prevalent primary brain tumor requiring maximal safe resection to significantly enhance survival depends on effective real- time histopathological input, which can be hampered by practical constraints such as distance of the laboratory and availability of neuropathology services. Over recent years the financial challenges have affected the effective running of these services in the National Health Service. This study evaluates the effectiveness of probe-based laser confocal endomicroscopy (p-CLE) for in-situ tumour margin assessment during surgery. <h3>Methods</h3> A systematic review and meta-analysis, registered with PROSPERO and guided by PRISMA, analyzed data from 1806 to 2022 across multiple databases, employing a random effects model. <h3>Results</h3> Out of 4182 studies screened, 6 were meta-analyzed, covering 336 samples using p-CLE. The sensitivity of p-CLE was substantial at 78.8%, while specificity was near perfect at 99.7%, with an overall accuracy of 82.5%. <h3>Discussion</h3> The study suggests that AI-enhanced p-CLE demonstrates high accuracy, sensitivity, and specificity in classifying glioblastoma, comparable to machine-learning algorithms and histopathologists. With an AUC of 86%, the potential of AI algorithms to differentiate tumour from non-tumor tissue indicates that future algorithmic improvements could enhance these outcomes. <h3>Conclusion</h3> A hybrid-federated approach incorporating these technologies is proposed to optimize surgical precision and patient outcomes by applying novel methods.