Noninvasive Intracranial Pressure Prediction Using a Multimodal Ultrasound-Based Hemispheric Modeling Strategy: A Prospective Dual-Center Study.

Background: Accurate intracranial pressure (ICP) surveillance is a cornerstone of neurocritical care management, yet invasive monitoring still depends on neurosurgical expertise, specialized hardware, and continuous bedside resources-factors that restrict universal use even though insertion-related complications are relatively uncommon. Contemporary noninvasive ultrasound methods have limited predictive accuracy and seldom incorporate affected side information. We therefore preliminarily validated a multimodal, ultrasound-based hemispheric modeling approach that blends hemodynamic and structural indexes while emphasizing affected side specificity to enhance ICP prediction.

Methods: In this prospective, dual-center study, 41 neurosurgical patients provided 216 paired ultrasound and invasive ICP measurements. Affected side and contralateral ultrasound parameters including pulsatility index, resistance index, optic nerve sheath diameter (ONSD), optic disk height, and ONSD-to-eyeball diameter ratio were analyzed. Linear mixed-effects models (LMMs) predicted continuous ICP, whereas generalized LMMs classified elevated ICP (≥ 20 mm Hg).

Results: Affected side parameters showed consistently stronger ICP correlations than unaffected side parameters. An affected side five-parameter LMM (pulsatility index, resistance index, ONSD, ONSD-to-eyeball diameter ratio, and optic disk height) provided superior continuous ICP prediction (coefficient of determination [R] = 0.618, root mean square error = 0.424), significantly outperforming contralateral models (R = 0.338, root mean square error = 0.558). For binary classification, affected side ONSD demonstrated excellent accuracy (area under the receiver operating characteristic curve = 0.927, sensitivity = 91.4%, specificity = 79.5%), whereas the optimal affected side seven-parameter generalized LMM reached an area under the receiver operating characteristic curve of 0.829 (sensitivity = 80.8%, specificity = 75.8%).

Conclusions: This study demonstrated the feasibility and potential advantages of a novel hemispheric (side-specific) modeling strategy for noninvasive ICP assessment. The multiparameter model constructed using affected side ultrasound parameters exhibited promising predictive accuracy, providing a potentially valuable and innovative noninvasive approach for ICP monitoring in neurocritical care patients that may serve as an adjunct when invasive monitoring is unavailable, although further validation is warranted.