A Real-Time Deep Learning Approach for Inferring Intracranial Pressure from Routinely Measured Extracranial Waveforms in the Intensive Care Unit

ObjectiveIntracranial pressure (ICP) is a physiological variable used to assess the neurological state of patients with life-threatening intracranial pathology, such as traumatic brain injury or stroke. The current standard of care for measuring ICP requires a catheter to be inserted into the brain, which is associated with an appreciable risk of hemorrhage and infection. We hypothesize that ICP can be computed from extracranial waveforms routinely measured in the Intensive Care Unit (ICU), such as invasive arterial blood pressure (ABP), photoplethysmography (PPG), and electrocardiography (ECG). MethodsWe extracted 600 hours of simultaneous ABP, ECG, PPG, and ICP data (sampled at 125 Hz) across 10 different patients from the MIMIC III Waveform Database. These recordings were segmented into 10 second windows and used to train six different deep learning models with ABP, ECG, and PPG waveforms as input features. Models were evaluated in both a singlepatient analysis and multi-patient analysis. ResultsThe performances of the six deep learning models were compared, revealing two tiers of performance. Among the top-tier models, the mean average error (MAE) for inferring ICP was approximately 1.50 mmHg for singlepatient analysis and 5 mmHg for multi-patient analysis. ConclusionsThese preliminary and novel results indicate the feasibility and accuracy of noninvasive ICP estimation by training deep learning models with extracranial physiological data. With further validation, this approach could be implemented in a continuous real-time fashion, thereby reducing risks associated with invasive monitoring and allowing more timely treatment of patients with critical brain injuries.