Feasibility comparison of deep learning image regressions to estimate intracranial pressure from cranial computed tomography in hydrocephalus

Objectives: Hydrocephalus (HCP) is a neurosurgical condition that causes functional disability. The purpose of the present study was to compare the predictive performances of various deep learning (DL) architectures in predicting intracranial pressure (ICP) values from cranial computed tomography (CT) scans. Materials and Methods: We retrospectively analyzed 242 HCP patients who underwent external ventricular drainage with intraoperative ICP measurement. Four clinically relevant axial CT slices were extracted per patient and used to train five DL architectures: Convolutional neural network (CNN), dense convolutional network, LeNet, and residual neural network. Performance was evaluated using mean absolute error (MAE), mean squared error (MSE), root MSE (RMSE), and coefficient of determination (R 2 ). Results: The best-performing model (CNN) achieved MAE 4.96 cmH2O and RMSE 6.22 cmH2O, but explained only a small proportion of variance (R 2 = 0.18). Model performance was modest across all architectures, with limited explanatory power. A contributing factor was the dataset imbalance, as a large proportion of patients presented with moderate to high ICP values, increasing variability and constraining regression learning across the full ICP spectrum. Conclusion: DL regression used to cranial CT demonstrated limited explanatory performance for estimating ICP and is not sufficient as a substitute for invasive monitoring. However, these findings provide preliminary feasibility evidence and a challenge for improvement. Future research using larger, balanced, multi-center datasets, and incorporating multimodal data will be necessary to enhance predictive accuracy and clinical applicability.