Background
Non-invasive measurement of intracranial pressure (ICP) has failed for decades. Yet, it would allow for proper evaluation of unconscious patients suffering from non-penetrating traumatic brain injury (TBI). Transcranial transmission ultrasound (TTUS) measurements provided promising experimental data via brain pulsatility. This study investigates its potential for the detection of elevated ICPs via machine learning-based analysis.
Methods
Patients with severe TBI and invasive ICP monitoring were prospectively enrolled in our intensive care unit. ICP, arterial blood pressure, heart rate and TTUS measurements were simultaneously recorded in situations with and without elevated ICP. A classification model was implemented based on measurements derived from 9 patients with 387 episodes of increased ICP (> 15 mmHg) and 345 episodes of normal ICP (< 10 mmHg). The model was validated in a leave-one-subject-out procedure.
Results
25 patients aged 61·6 ± 17·6 years were enrolled from October 2021 to October 2022. 279 data sets with a mean ICP of 11·3 mmHg (1st quartile 6·1 mmHg; 3rd quartile 14·8 mmHg) were acquired and analyzed. Automated analysis of the TTUS measurements successfully identified increased ICP values > 15 mmHg with a sensitivity 100% and a specificity 47%. A negative predictive value of 100% was achieved, the positive predictive value was 14% for the test set.
Conclusions
TTUS can precisely exclude elevated ICP in TBI patients with a negative predictive value of 100%. Despite low specificity, exclusion of raised ICP can already partially identify patients in the field requiring immediate imaging and potentially neurosurgical intervention. Worth mentioning, this is the very first approach achieving such a high reliability.