Brain tumors detection and segmentation in MR images: Gabor wavelet vs. statistical features

Object The aim of this study was to evaluate the integration of positron emission tomography (PET) scanning data into the image-guided resection of brain tumors. Methods Positron emission tomography scans obtained using fluorine-18 fluorodeoxyglucose (FDG) and l-[methyl-11C]methionine (MET) were combined with magnetic resonance (MR) images in the navigational planning of 103 resections of brain tumors (63 low-grade gliomas [LGGs] and 40 high-grade gliomas [HGGs]). These procedures were performed in 91 patients (57 males and 34 females) in whom tumor boundaries could not be accurately identified on MR images for navigation-based resection. The level and distribution of PET tracer uptake in the tumor were analyzed to define the lesion contours, which in turn yielded a PET volume. The PET scanning–demonstrated lesion volume was subsequently projected onto MR images and compared with MR imaging data (MR volume) to define a final target volume for navigation-based resection—the tumor contours were displayed in the microscope’s eyepiece. Maximal tumor resection was accomplished in each case, with the intention of removing the entire area of abnormal metabolic activity visualized during surgical planning. Early postoperative MR imaging and PET scanning studies were performed to assess the quality of tumor resection. Both pre- and postoperative analyses of MR and PET images revealed whether integrating PET data into the navigational planning contributed to improved tumor volume definition and tumor resection. Metabolic information on tumor heterogeneity or extent was useful in planning the surgery. In 83 (80%) of 103 procedures, PET studies contributed to defining a final target volume different from that obtained with MR imaging alone. Furthermore, FDG-PET scanning, which was performed in a majority of HGG cases, showed that PET volume was less extended than the MR volume in 16 of 21 cases and contributed to targeting the resection to the hypermetabolic (anaplastic) area in 11 (69%) of 16 cases. Performed in 59 LGG cases and 23 HGG cases, MET-PET demonstrated that the PET volume did not match the MR volume and improved the tumor volume definition in 52 (88%) of 59 and 18 (78%) of 23, respectively. Total resection of the area of increased PET tracer uptake was achieved in 54 (52%) of 103 procedures. Conclusions Imaging guidance with PET scanning provided independent and complementary information that helped to assess tumor extent and plan tumor resection better than with MR imaging guidance alone. The PET scanning guidance could help increase the amount of tumor removed and target image-guided resection to tumor portions that represent the highest evolving potential.

Download Full-text

LATERAL VENTRICLE TEXTURE ANALYSIS IN ALZHEIMER BRAIN MR IMAGES USING KERNEL DENSITY ESTIMATION

Biomedical Sciences Instrumentation ◽

10.34107/yhpn9422.04241 ◽

2021 ◽

Vol 57 (2) ◽

pp. 241-246

Author(s):

Deboleena Saddhukhan ◽

◽

Amrutha Veluppal ◽

Anandh Kilpattu Ramaniharan ◽

Ramakrishnan Swaminathan ◽

...

Keyword(s):

Density Estimation ◽

Kernel Density Estimation ◽

Neurodegenerative Disorder ◽

Large Population ◽

Kernel Density ◽

Svm Classifier ◽

Statistical Features ◽

Mr Images ◽

Brain Mr Images ◽

Textural Changes

Alzheimer's Disease (AD) is an irreversible, progressive neurodegenerative disorder affecting a large population worldwide. Automated diagnosis of AD using Magnetic Resonance (MR) imaging-based biomarkers plays a crucial role in disease management. Compositional changes in cerebrospinal fluid due to AD might induce textural variations in Lateral Ventricles (LV) of the brain. In this work, an attempt has been made to differentiate Alzheimer's condition by quantifying the textural changes in LV using Kernel Density Estimation (KDE) technique. Reaction-Diffusion level set method is used to segment the LV from T1-weighted trans-axial brain MR images obtained from a publicly available database. Spatial KDE is used to analyze the local intensity variations within the segmented LV. The optimal kernel function and bandwidth are selected for KDE. The statistical features such as mean, median, standard deviation, variance, kurtosis, skewness and entropy, representing the distribution of KDE values within LV, are evaluated. The extracted KDE-based statistical features show significant discrimination between normal and AD subjects (p<0.01). An accuracy of 86.20% and sensitivity of 96% are obtained using SVM classifier. The results indicate that KDE seems to be a potential tool for analyzing the textural changes in brain, and thus can be clinically relevant for diagnosis of AD.

Download Full-text