(Dr. David N. Ku, advisor)
"Knowledge-Based Magnetic Resonance Angiography"
Conventional x-ray angiography, the primary means of quantifying cardiovascular disease, is invasive and expensive. Magnetic resonance angiography (MRA) is a noninvasive and less expensive alternative to conventional angiography. Unfortunately, the current state of MRA has poor accuracy in quantifying vascular disease because turbulent and accelerating blood flow in diseased arteries induces signal loss. Most investigators have attempted to eliminate signal loss by developing faster, more powerful scanners and imaging sequences; however, signal loss is still a limitation despite these advances.
This current work takes a different approach: Signal loss may be considered a signature--sufficient to identify the geometry of the diseased artery. The goal of this study was to extract the percent stenosis, a major criterion for deciding patient treatment, from measurable attributes of post stenotic signal loss.
Analytical and experimental investigations of post-stenotic signal loss identified several parameters in the MR angiogram which were strongly related to the stenosis geometry. A neural network based expert system was developed to find the relationship between the signal loss parameters and the degree of narrowing. Phantom studies showed that as few as seven parameters were necessary to extract the percent stenosis with 97% accuracy.
A proof-of-concept study was performed on a series of 21 patients (42 carotid arteries) with clinical symptoms. Patients were given both an MRI and x-ray (the gold standard). The x-ray angiograms were the basis of comparison in this study. The 42 arteries were divided into separate training and testing groups. The expert system extracted the severity of artery narrowing with 94% accuracy and a 0.98 correlation coefficient. The worst case had a 9% measurement error. It is hoped that the technique of treating signal loss as information will be a useful diagnostic aid and will increase the use of non-invasive MRI technology to diagnose cardiovascular disease.