NSF Award
1556110
The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase II project will be to improve the quality and decrease the high costs associated with treating patients who suffer severe traumatic brain injuries. This project aims to develop an accurate, affordable (<$100 per use) and non-invasive device to monitor a patient's intracranial pressure following head injury. Increased intracranial pressure can result in poor health outcomes including long-term disability or death, if left untreated. However, the only available method to monitor intracranial pressure is expensive (~$10,000 per patient) and requires neurosurgery. The lack of a method to accurately screen patients to determine who needs surgery results in misdiagnoses and incorrect treatment in about 46% of patients among an estimated 50,000 patients in the US alone, and hundreds of thousands more globally. Successful commercialization of product is expected to result in savings in the range $250 million ever year to the US healthcare system.<br/><br/>The proposed project will develop a medical device to accurately display a patient's intracranial pressure non-invasively and for use outside of the neurocritical care unit. The core technological approach of the proposed work is the analysis of blood flow velocity waveforms using advanced signal processing methods in a machine-learning framework. The machine-learning framework allows experience-based learning utilizing prior, established databases of waveforms that have been well-characterized. Three new machine-learning paradigms that utilize the shape features of the blood flow velocity waveforms will be utilized to progressively increase accuracy of intracranial pressure estimation. The first will establish a basic estimate using shape features of individual waveform pulses, considered independent of neighboring pulses. Subsequently, clinically established features of the waveform will be utilized to learn causal changes in the shape features resulting from changes in intracranial pressure. Finally, the shape features in successive pulses will be used as a sequence to machine-learn the intracranial pressure estimate. Together, these will enable increased accuracy in estimation. All of the methods proposed in this program are entirely novel. This approach allows for real time monitoring at an affordable price point that is within current reimbursement limits for ultrasonography procedures.
