Predicting Intracranial Pressure Levels: A Deep Learning Approach Using Computed Tomography Brain Scans.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Elevated intracranial pressure (ICP) is a critical condition requiring rapid diagnosis to prevent severe neurological damage.
• Invasive ICP measurement is the gold standard but is time-consuming and carries risks.
• Existing noninvasive methods are often experimental, and AI has not fully utilized CT scans for ICP evaluation due to data limitations.

Purpose of the Study:

• To address the gap in AI-driven ICP evaluation from CT scans.
• To develop and train deep learning models using a custom dataset incorporating demographic and Glasgow Coma Scale (GCS) data.
• To classify ICP levels, predicting whether they exceed a threshold of 15 mm Hg.

Main Methods:

• Developed four distinct deep learning models.
• Utilized a custom dataset of 578 paired CT brain scans with ICP values, GCS scores, and demographic data.
• Incorporated demographic and GCS data as additional input channels for the models.
• Trained models for a binary classification task to predict ICP above 15 mm Hg.

Main Results:

• The best-performing models achieved an area under the curve of 88.3% and a recall of 81.8%.
• An explainability algorithm provided insights into model decision-making processes.
• Demonstrated the models' ability to focus on relevant areas within CT scans.

Conclusions:

• AI models show significant potential for evaluating ICP from brain CT scans with high recall.
• The study highlights the feasibility of using AI for faster ICP assessment.
• Further research is needed to validate findings and enhance clinical applicability.