Arterial input function segmentation based on a contour geodesic model for tissue at risk identification in ischemic stroke

Abstract

Perfusion parameters such as cerebral blood flow (CBF) and T max have been proven to be useful in the diagnosis and prognosis for ischemic stroke. Arterial input function (AIF) is required as an input to estimate perfusion parameters. This makes the AIF selection paradigm of clinical importance. Methods: This study proposes a new technique to address the problem of AIF selection, based on a variational segmentation model that combines geometric constraint in a distance function. The modified model uses discrete total variation in the distance term and via minimizing energy locates the arterial regions. Matrix analysis is utilized to identify the AIF with maximum peak height within the segmented region. Results: Group mean differences indicate that overall the AIF selected by the purposed method has better arterial features of higher peak position (16.7 and 26.1 a.u.) and fast attenuation (1.08 s and 0.9 s) as compared to the other stateof -the-art methods. Utilizing the selected AIF, mean CBF, and T max values were estimated higher than the traditional methods. Ischemic regions were precisely located through the perfusion maps. Conclusions: This AIF segmentation framework worked on perfusion images at levels superior to the current clinical state of the art. Consequently, the perfusion parameters derived from AIF selected by the purposed method were more accurate and reliable. The proposed method could potentially be considered as part of the calculation for perfusion imaging in general.

Core and penumbra estimation using deep learning-based AIF in association with clinical measures in computed tomography perfusion (CTP)

“…Traditional AIF calculation is done either manually or by using automatic methods such as clustering and arterial likelihood methods [ 11 , 30 ]. For manual AIF selection, a trained clinician operator based on his experience selects a small number of pixels on brain image as AIF [ 11 , 29 ].…”

“…Clustering algorithm estimated three AIFs for each dataset in the training and validation sets to generate training and validation AIF curves. To estimate the AIF, this clustering algorithm employs recursive cluster analysis in the middle cerebral artery (MCA) region [ 29 , 30 ]. Training dataset density was improved for model training performance using feature augmentation and spline interpolation.…”