Explore the vast range of titles available.

ObjectivesTo investigate whether utilizing a convolutional neural network (CNN)-based arterial input function (AIF) improves the volumetric estimation of core and penumbra in association with clinical measures in stroke patients.MethodsThe study included 160 acute ischemic stroke patients (male = 87, female = 73, median age = 73 years) with approval from the institutional review board. The patients had undergone CTP imaging, NIHSS and ASPECTS grading. convolutional neural network (CNN) model was trained to fit a raw AIF curve to a gamma variate function. CNN AIF was utilized to estimate the core and penumbra volumes which were further validated with clinical scores. ResultsPenumbra estimated by CNN AIF correlated positively with the NIHSS score (r = 0.69; p < 0.001) and negatively with the ASPECTS (r =  − 0.43; p < 0.001). The CNN AIF estimated penumbra and core volume matching the patient symptoms, typically in patients with higher NIHSS (> 20) and lower ASPECT score (< 5). In group analysis, the median CBF < 20%, CBF < 30%, rCBF < 38%, Tmax > 10 s, Tmax > 10 s volumes were statistically significantly higher (p < .05).ConclusionsWith inclusion of the CNN AIF in perfusion imaging pipeline, penumbra and core estimations are more reliable as they correlate with scores representing neurological deficits in stroke. Critical relevance statementWith CNN AIF perfusion imaging pipeline, penumbra and core estimations are more reliable as they correlate with scores representing neurological deficits in stroke.Graphic abstractKey pointsCNN-based AIF improves the estimation of penumbra and infarct core volumes.CNN AIF identifies patients with core regions which were ignored by conventional approaches.Core and penumbra assessment with CNN AIF correlates well with clinical scores.

Background: Diagnosis and timely treatment of ischemic stroke depends on the fast and accurate quantification of perfusion parameters. Arterial input function (AIF) describes contrast agent concentration over time as it enters the brain through the brain feeding artery. AIF is the central quantity required to estimate perfusion parameters. Inaccurate and distorted AIF, due to partial volume effects (PVE), would lead to inaccurate quantification of perfusion parameters. Methods: Fifteen patients suffering from stroke underwent perfusion MRI imaging at the Tri-Service General Hospital, Taipei. Various degrees of the PVE were induced on the AIF and subsequently corrected using rescaling methods. Results: Rescaled AIFs match the exact reference AIF curve either at peak height or at tail. Inaccurate estimation of CBF values estimated from non-rescaled AIFs increase with increasing PVE. Rescaling of the AIF using all three approaches resulted in reduced deviation of CBF values from the reference CBF values. In most cases, CBF map generated by rescaled AIF approaches show increased CBF and Tmax values on the slices in the left and right hemispheres. Conclusion: Rescaling AIF by VOF approach seems to be a robust and adaptable approach for correction of the PVE-affected multivoxel AIF. Utilizing an AIF scaling approach leads to more reasonable absolute perfusion parameter values, represented by the increased mean CBF/Tmax values and CBF/Tmax images.

Perfusion-weighted imaging (PWI) is a noninvasive Magnetic Resonance (MR)/Computed Tomography (CT) technique that assesses various hemodynamic parameters to examine blood flow in brain regions. These parameters are used in stroke patients to locate the penumbra or the tissue at risk, which can be salvaged with reperfusion therapies. Certain artefacts are associated with the imaging pipeline and segmentation methods used in scanner softwares to estimate perfusion parameters. The focus of this thesis is the development of an image analysis pipeline for perfusion parameter estimation and segmentation models. We concentrate on difficult problems such as arterial region segmentation on perfusion weighted images and tumor segmentation on images with low contrast, in-homogeneous intensity, and non-smooth edges.We begin with developing a arterial region segmentation model in the variational framework. We propose a new model in which geometric constraints are incorporated into a distance function. The modified model employs discrete total variation in the distance term and locates arterial regions by minimizing the energy of a convex functional, outperforming previous selective segmentation works that typically employ either a cost function or a clustering-based approach. This enhancement enables our model to effectively select an arterial region that performs well in identifying tissue at risk. Another work investigates whether fitting a hemodynamic model to the Arterial input function (AIF) obtained from arterial segmentation and minimizing the partial volume effect during AIF selection improves volumetric estimation of core and penumbra in stroke patients.In the second half of this thesis, we propose an efficient framework for selective segmentation using a new region force term and a geodesic distance penalty based on a discrete TV formulation. The proposed model is user-independent and allows for precise segmentation in tumor images and medical images with non-homogeneous, non-smooth, and scraggy boundary edges. A chapter is dedicated to integrate the variational segmentation method with deep learning. Despite being extremely popular recently, deep learning techniques are frequently constrained by the need for sizable sets of labelled data. We demonstrate how labels can be supplemented by using a variational method as a loss function in a unsupervised training algorithm for brain tumour segmentation.