Meeting Banner
Abstract #3493

Machine Learning-Based Cerebral Blood Flow Quantification for ASL MRI

Ze Wang1, Anna Rose Childress1, John A. Detre2

1Psychiatry, University of Pennsylvania, Philadelphia, PA, United States; 2Neurology, University of Pennsylvania, Philadelphia, PA, United States

Arterial spin labeling (ASL) is not stable across time but no one has taken this into account during perfusion quantification. Due to the systematic labeling and control labeling, ASL CBF quantification is a natural two-class data classification process. Based on this phenomenon, we used a powerful machine learning algorithm, the support vector machine (SVM), to extract the spin labeling function from the ASL data and used it for CBF quantification. The method demonstrated significantly improved temporal SNR and spatial image quality for CBF quantification using normal healthy subjects data and data from patients with Alzheimers Disease.

Abstract PDF E-Poster Video

Keywords

accordingly almost amount apart approved arrows arterial assess assumed bars better blood body bottom brain calculation calculations caused cerebral checked classification classifier classify cleaned collection color consent consider constant continuous control correlated correlations courses curve delay described dimension discovery disease distance divided evaluate extract extracted feature fixed flow fluctuations forms function grants hardware head healthy ideal identified illustrated illustration imperfections improve improved improvement indicated intuitively knowledge label labeled labeling learning likely linear linearly located machine mapped maps marked markedly materials matrix middle mining months motion motions much neither neurology nonlinear novel oscillating ovals pairs paradigm part patients polynomial post practice preprocessed previously procedures process proposed psychiatry published quantification real regressed regular remarkable removal removing residuals respectively rose sample save scalar scanner separable separated separating sequential series signed significantly simplicity slightly space spatial spin spins subject subjects subtle suggesting summary support supported though transform twice variation variations vector vectors water whole yielded