SFMR:s företagsstipendium: Woosung Seo

Manual and automatic quantification of ventricular CSF volume in iNPH patients undergoing shunt surgery

INTRODUCTION: Neuroimaging is essential in the work-up of idiopathic normal pressure hydrocephalus (iNPH). This study aims to compare manual segmentations with fully automated quantifications of lateral ventricular volume on CT brain images. Both preoperative and follow-up scans (6 weeks postop) were analyzed, and results were compared to clinical response.

METHODS: Forty-nine iNPH patients were included. CT images were analyzed before and after shunt surgery using manual segmentations and ScanDx, an AI-driven brain CT quantification pipeline. It uses trained deep-learning models to segment and quantify various brain tissue classes, including ventricles from CT images. Clinical assessment was performed before and 3 months after shunt insertion using a modified iNPH symptom scale, including 10m gait, balance and incontinence evaluation, and MMSE. Responders were predefined as having either ³ 3 points improvement on MMSE, ³ 5 points on iNPH scale, or both.

RESULTS: Manual and automatic volumetry of lateral ventricles were strongly correlated (Pearson’s r=0.96 for preoperative and r=0.91 for postoperative volumes. Median volume reduction at 6 weeks was 21.5 ml (15.2%) for automatic measurements and 13.5ml (12.4%) for manual. Clinical response was observed in 30 of 42 (71%) patients but there was no correlation between clinical improvement and degree of VCSF volume reduction.

CONCLUSION: Our study demonstrates the potential of acquiring quantitative volumetric data from CT images, using a fully automated AI tool, even with the presence of an intraventricular shunt. Robust correlations were seen between automatic and manual methods. However, a minor systematic bias was observed where the automatic model tends to overestimate certain anatomic regions.