SBIR Phase I: Development of a Novel Diagnostic Test for Pulmonary Embolism Based on Artificial Intelligence and Spectral Analysis of Blood

点击翻译按钮获取中文摘要

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will result from the development of a fast, non-invasive, and highly accurate test to diagnose pulmonary embolism in the emergency department. In the United States, pulmonary embolism (PE) affects up to 1 million patients per year and is responsible for nearly 100,000 yearly deaths. Its diagnosis is challenging due to the presentation of nonspecific symptoms and the lack of high-accuracy screening methods. While the current standard of care is to rule out PE with an established blood test (D-Dimer), approximately 90% of those results are false positives, causing the test to be used with restraint in the clinic, and leading to both the underdiagnosis of the disease and the overuse of strongly radiative imaging methods like CT pulmonary angiograms. A new, highly specific test for PE could increase patient safety, standardize clinical care processes, reduce costs and save lives.This Small Business Innovation Research (SBIR) Phase I project will develop and validate a new diagnostic tool for PE based on the combination of fast blood spectroscopy and modern machine learning (ML) algorithms. A key aim of the research is demonstrating that ML combined with blood spectroscopy can substantially outperform the D-Dimer biomarker test, which has notoriously low specificity (~40%). An important Phase I milestone will be to show that the specificity of the resulting PE test either (a) already surpasses that of the D-Dimer test when trained on the relatively small dataset used in this Phase I proposal, or (b) substantially increases with the size of the training dataset, so that the test can outperform D-Dimer simply by procuring a larger pool of blood samples. The technical challenges addressed in this phase include evaluating different spectroscopic methods and modalities, minimizing the coefficient of variation for spectra acquisition, as well as designing and optimizing ML models for one-dimensional spectral data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.