Illuminating the black box: Non-invasive intracranial pressure estimation via near-infrared photonics and deep learning
Abstract
Continuous intracranial pressure (ICP) monitoring is essential for managing severe neurological conditions; however, traditional methods require invasive surgical skull drilling, which carries significant risks of infection and hemorrhage. This presentation introduces the SafeICP platform, a non-invasive alternative that measures microvascular cerebral blood flow at the bedside using speckle contrast optical spectroscopy and near-infrared biophotonics. A major focus will be the project's deep learning framework, demonstrating how advanced architectures – specifically InceptionTime and Multi-Wavelet Decomposition Network – can capture subtle temporal dynamics in raw blood flow index time-series and translate them directly into absolute ICP values without manual feature extraction. We will also present a complementary feature-based machine learning approach that leverages pulse morphology descriptors and patient demographics through gradient boosting and random forest models.