Abstract
<title>Abstract</title> <p>Intracranial compliance determines how intracranial pressure (ICP) responds to additional intracranial volume, but the pressure--volume curve is not directly available during routine bedside monitoring. Entropy measures of the ICP waveform may carry information about this hidden mechanical state. We performed a secondary analysis of a previously published porcine model of reversible intracranial hypertension to test whether entropy-derived latent states recover experimental pressure--volume structure. ICP was acquired at 200 Hz. Normalized permutation entropy, normalized wavelet entropy and sample entropy were computed in 2 s windows with 1.5 s overlap, producing one feature vector every 0.5 s. A three-state Gaussian hidden Markov model was fitted using only entropy-derived features; absolute ICP, infused volume, pressure--volume slope and threshold labels were not model inputs. After inference, states were projected onto the ICP--volume trajectory and labelled Normal, Warning and Alert according to the corresponding pressure--volume slope. State-specific dP/dV, an inverse marker of compliance, increased across these states: 0.53, 3.84 and 7.96 mmHg ml$^{-1}$. Leave-one-subject-out validation yielded comparable low-, intermediate- and high-slope regimes. Thus, pressure--volume structure can be recovered from ICP waveform complexity without measuring volume or compliance directly. This preclinical proof of concept requires validation in larger experimental and clinical datasets and should not be interpreted as a clinically validated alarm system.</p>