Advancements in Artificial Intelligence for Hydrocephalus Diagnosis: A Systematic Review

Introduction: Artificial intelligence (AI), specifically machine learning (ML), has shown promise in advancing diagnostic accuracy for hydrocephalus in both pediatric and adult populations. Hydrocephalus diagnosis, which typically relies on imaging modalities, may benefit from the integration of ML, particularly in tasks requiring precise volumetric and morphological assessments.

Methods: We conducted a systematic review across multiple databases, including MEDLINE, Embase, and Web of Science, up to January 2024. A total of 384 records were identified, with 135 duplicates removed and 216 articles excluded after screening. Following a rigorous eligibility assessment, 33 studies were included in the final analysis, focusing on AI applications in hydrocephalus diagnostics.

Results: Our systematic review included 33 studies, with 51.52% focusing on normal pressure hydrocephalus, 27.27% on pediatric hydrocephalus, and 21.21% on other types. A total of 45 ML models were analyzed, with deep learning models being the most prevalent (51.11%), followed by traditional ML (28.89%) and hybrid models (8.89%). The majority of studies used MRI (62.5%) and CT (27.5%) as input modalities, with supervised learning employed in 84.44% of models. The ML tasks included volumetry (34%), morphology (32%), and segmentation (26%), with smaller proportions dedicated to parcelation and CSF pressure patterns (4% each). The median diagnostic accuracy across models was 90%, with an area under the curve (AUC) of 0.94 and a Dice coefficient of 0.93.

Conclusion : This review highlights the potential of AI in improving hydrocephalus diagnostics, demonstrating high accuracy and efficiency in various tasks. However, challenges remain in clinical adoption due to unfamiliarity among practitioners. Future studies should aim to validate these findings in real-world settings to determine the utility of AI in enhancing hydrocephalus patient management and addressing challenges in resource-limited environments.