Integrate Dynamic System Model and Machine Learning for Calibration-Free Noninvasive ICP

点击翻译按钮获取中文摘要

Project Summary No clinical device exists for noninvasive intracranial pressure (nICP) assessment. Past attempts have focused on identifying ICP-related signals that are noninvasively measureable, but have done little to address the calibration problem. Without calibration, only ICP trending can be inferred at the best. However, noninvasive calibration is not trivial. A universal calibration will fail because individual patients require different calibration to obtain accurate results. On the other hand, the use of plain regression for individualized calibration is infeasible because ICP cannot be obtained noninvasively for a de novo patient to begin with. Invasive ICP monitoring remains a standard of care and this can be leveraged to continuously grow a database of ICP, noninvasive signals, and different calibration equations, e.g., each built from a pair of invasive ICP and noninvasive signal in the database. Then nICP becomes feasible by selecting from a rich set of calibration equations the optimal choice for a de novo patient. In this project, we will pursue three aims that will lead to the development of an accurate noninvasive ICP system based on Transcranial Doppler. These aims are: 1) To implement and validate core algorithms needed for achieving accurate nICP; 2) To test if estimated nICP is sensitive to variations in ultrasound probe placement; 3) To test the generalizability of the proposed nICP approach. Large epidemiologic surveys reveal that ICP is monitored in only about 58% of US patients when ICP monitoring is indicated. It is a smaller percentage (37%) in European patients and even fewer in developing countries. The proposed nICP approach does not have the high risks associated with invasive ICP, requires no onsite neurosurgical expertise, and can be economically deployed and readily practiced. Therefore, its potential impact is enormous.