The fully automated and relatively accurate method of brain tissues segmentation on T2-weighted magnetic resonance perfusion images is proposed. Segmentation with this method provides a possibility to obtain perfusion region of interest on images with abnormal brain anatomy that is very important for perfusion analysis. In the proposed method the result is presented as a binary mask, which marks two regions: brain tissues pixels with unity values and skull, extracranial soft tissue and background pixels with zero values. The binary mask is produced based on the location of boundary between two studied regions. Each boundary point is detected with CUSUM filter as a change point for iteratively accumulated points at time of moving on a sinusoidal-like path along the boundary from one region to another. The evaluation results for 20 clinical cases showed that proposed segmentation method could significantly reduce the time and efforts required to obtain desirable results for perfusion region of interest detection on T2-weighted magnetic resonance perfusion images with abnormal brain anatomy.
Deep Dive into Brain Tissues Segmentation on MR Perfusion Images Using CUSUM Filter for Boundary Pixels.
The fully automated and relatively accurate method of brain tissues segmentation on T2-weighted magnetic resonance perfusion images is proposed. Segmentation with this method provides a possibility to obtain perfusion region of interest on images with abnormal brain anatomy that is very important for perfusion analysis. In the proposed method the result is presented as a binary mask, which marks two regions: brain tissues pixels with unity values and skull, extracranial soft tissue and background pixels with zero values. The binary mask is produced based on the location of boundary between two studied regions. Each boundary point is detected with CUSUM filter as a change point for iteratively accumulated points at time of moving on a sinusoidal-like path along the boundary from one region to another. The evaluation results for 20 clinical cases showed that proposed segmentation method could significantly reduce the time and efforts required to obtain desirable results for perfusion reg
Dynamic susceptibility contrast (DSC) perfusion magnetic resonance (MR) imaging has already been widely used for the management of patients with brain tumors and cerebrovascular diseases, such as vascular stenosis or stroke [1][2][3]. It is based on the fact that T2-weighted MR images show decreased signal intensity while the iodinated contrast agent passes through the tissue. DSC exam output is a series of T2 weighted images acquired before, during, and after iodinated contrast agent injection into the vascular system. This time-sequence data reflect local changes in perfusion tissue characteristics. To estimate perfusion tissue characteristics according to the obtained time series it is necessary to carry out analysis of signal intensity changes on a pixel-by-pixel basis within the same part of the human body.
The results of DSC perfusion exam can be used both for quantitative estimation of perfusion characteristics as well as visual assessment of tissue perfusion. It is visualization of perfusion characteristics on the so-called perfusion maps that serves to detect regions with potential perfusion lesions and determine a diagnosis. However, brain lesions can be poorly visualized on perfusion maps due to the low contrast between the lesion and surrounding tissues. Such effect occurs when perfusion maps display extremely high values for noised pixels or skull and extracranial soft tissues regions. Thus, the important step of DSC perfusion analysis is pre-processing of time-sequence data through brain tissues segmentation and binary mask creation for so-called perfusion region of interest (ROI) [4][5][6][7].
manual segmentation gives more accurate results it is more laborious and time-consuming. It should be mentioned that to obtain accurate segmentation results an operator has to have sufficient knowledge and experience to identify brain anatomical structures and its lesions on MR images. However, regarding the disadvantages of manual segmentation mentioned here, the automated segmentation is not that ideal for this task [6,7]. In order to be of practical use, the segmentation method should provide accurate results for images with abnormal brain anatomy. It is automated segmentation methods that generally give wrong perfusion ROI for such images, as a result, segmentation faults lead to the absence of perfusion characteristics data for pixels in regions with potential perfusion lesions. Therefore, there are a number of different methods of brain tissues segmentation on MR images [8][9][10][11][12].
Most of the automated brain tissues segmentation methods are used for T1-weighted MR images processing. For clarity, these methods can be divided into two groups: those which are based on pixels intensity analysis (thresholding and clustering algorithms), and those which use patterns (neural network classifiers and atlas-based algorithms). The central problem of the first group methods is overlapping pixel intensity values in region of interest and background. The second group problem is lack of age-sex-race-specific pre-segmented template data and lack of training samples for different size, density, and volume of brain lesions.
Surrounding brain tissues appear as bright pixels on T2-weighted MR perfusion images. It is the main reason for segmentation faults while using automated methods which work fine with T1weighted images [13]. The proposed strategy of parameter-based transformation of Т2-weighted into Т1-weighted image intensities [14] solves partially the issue of using suitable for T1-weighted images methods to segment T2-weighted images. However, it does not solve the problem of segmentation faults for images with abnormal brain anatomy.
The idea of using CUSUM filter to track the boundary for autonomous vehicles [15,16] was applied to solve image segmentation issue in general [17]. The main purpose of CUSUM filter usage is to solve the issue of boundary tracking at time of movement between two regions using only local information in the presence of noise. Thus, it gives the opportunity to solve the segmentation issues in different domains [18,19]. But in order to get accurate segmentation results using CUSUM filter in specific domain relationship between two disparate regions should be established and appropriate movements pattern should be developed so that the boundary can be accurately tracked and estimated [17,18].
The aim of the current study is to develop a fully automated method of brain tissues segmentation on Т2-weighted MR perfusion images of a human head with abnormal brain anatomy using CUSUM filter for boundary pixels.
To achieve the aim, the following tasks have been set: ̶ to develop an algorithm for path planning with specific movements pattern along the boundary between two regions: one region is specified with brain tissues pixels and the other one with skull, extracranial soft tissues and background pixels; ̶ to adapt CUSUM filter for a change point detection in order to deter
…(Full text truncated)…
This content is AI-processed based on ArXiv data.
