China
Africa
Explore chapters and articles related to this topic
Machine Learning Applications In Medical Image Processing
Published in Sanjay Saxena, Sudip Paul, High-Performance Medical Image Processing, 2022
Tanmay Nath, Martin A. Lindquist
Finally, there are multiple free open-source software packages used for medical image processing. The most popular package for neuroimaging (including fMRI, PET, SPECT, EEG, and MEG processing) is statistical parametric mapping (SPM) [59], a MATLAB based toolbox that allows for end-to-end analysis. AFNI (analysis of functional neuroimages) [60] is another comprehensive package for analysis of anatomical and function MRI. It has many built-in functions written in C, Python, R, and shell scripts. Its installation files are zipped in a binary package which can be easily installed in any operating system (OS). FSL (FMRIB software library) [61] is another package that can be used to analyze fMRI, MRI, and DTI data. It is also written in C and consists of a series of programs for pre-processing, conducting statistical analysis, and visualizing the results. FreeSurfer [62] is another C based software package that studies the surface of brain, especially the cortical and sub-cortical anatomy using structural, functional MRI, DTI, and PET. 3D Slicer [63] is an open-source platform for medical image analysis (e.g., image registration and segmentation) across multiple modalities including MRI, CT, Ultrasound, nuclear medicine, and microscopy. Furthermore, there is a python-based pipeline namely Nipype (Neuroimaging in Python Pipelines and Interfaces) [64], which provides an interface to the above packages within a single framework. Thus, a user in Nipy (https://nipy.org/) can interactively explore algorithms from different packages and combine them to process data faster.
Introduction
Published in Atsushi Kawaguchi, Multivariate Analysis for Neuroimaging Data, 2021
SPM (Statistical Parametric Mapping, http://www.fil.ion.ucl.ac.uk/spm/) is the most widely used software in MATLAB®, see Friston et al. (2007) and (Ashburner, 2012) for details. MATLAB which is required for SPM is commercial, but adding SPM is free. A number of toolboxes (extensions) are available, especially VBM8 and CAT12, which are useful for morphological analysis of brain images. Analysis of SPECT and PET images is also available. In this book, statistical methods used in SPM are introduced as common methods for brain imaging analysis. FSL (http://www.fmrib.ox.ac.uk/fsl/) is a Linux-based software developed by the analysis group at Oxford University, and is widely used in the same way as SPM. It is especially suitable for the analysis of images showing the degree of water diffusion in the brain. See Smith et al. (2004), Woolrich et al. (2009) and Jenkinson et al. (2012) for details. FreeSurfer (http://surfer.nmr.mgh.harvard.edu/) is often used to calculate brain volume from cortical thickness and anatomical segments (Fischl, 2012). Since many processes can be performed at once with simple commands, the usage in structural brain analysis is rapidly increasing. ANTs (advanced normalization tools, http://stnava.github.io/ANTs/) can perform advanced techniques for image transformation (Avants et al., 2009). Furthermore, various analysis methods have been incorporated in the preprocessing process, which are useful for brain imaging analysis. R packages have also been developed for each of these software programs, which allow users to operate them from R and visualize the analysis output.
Alzheimer's Disease Classification
Published in Amitoj Singh, Vinay Kukreja, Taghi Javdani Gandomani, Machine Learning for Edge Computing, 2023
Monika Sethi, Sachin Ahuja, Vinay Kukreja
To preprocess the data, past researchers employed software such as Statistical Parametric Mapping (SPM) and FreeSurfer Library (FSL). FSL helps for brain extraction and tissue segmentation, whereas SPM helps align, spatially normalize, and smooth the brain scans. FreeSurfer has a pre-processing stream that involves three things, namely, skull stripping, non-linear registration, and segmentation. In [12], the researcher groups utilized a Clinica software framework proposed by ARAMIS Lab that facilitates the working of FSL, SPM, and FreeSurfer.
Motor Network Reorganization Induced in Chronic Stroke Patients with the Use of a Contralesionally-Controlled Brain Computer Interface
Published in Brain-Computer Interfaces, 2022
Joseph B. Humphries, Daniela J. S. Mattos, Jerrel Rutlin, Andy G. S. Daniel, Kathleen Rybczynski, Theresa Notestine, Joshua S. Shimony, Harold Burton, Alexandre Carter, Eric C. Leuthardt
A previously described pipeline preprocessed all functional MRI data [44]. The 4dfp suite (4dfp.readthedocs.io) of preprocessing steps comprised slice-time correction, removal of odd-even slice intensity differences, rigid body motion correction, affine transformation to a (3 mm)3 atlas space, spatial smoothing with a 6 mm FWHM Gaussian kernel, voxelwise linear detrending, and a temporal low pass filter (0.1 Hz cutoff). Freesurfer software performed cortical surface segmentation. Regression of nuisance waveforms, derived from motion correction timeseries, CSF signal, white matter signal, and the whole brain (‘global’) signal, reduced spurious variance [45,46]. High-motion frames were removed from the analysis [44]. Fisher z-transforms were applied to Pearson correlation coefficients prior to statistical analysis.
An automatic level set method for hippocampus segmentation in MR images
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2020
Nazanin Safavian, Seyed Amir Hossein Batouli, Mohammad Ali Oghabian
It is worth noting that we compared our results with Freesurfer because this package is a benchmark and widely used software, developed to segment cortical and subcortical brain structures using multiple segmentation frameworks. To compare our results with those obtained from Freesurfer, we used a number of statistical approaches, with the manual hippocampus segmentation being the gold standard. The average of Dice and Jaccard indices in both left and right hippocampus for our method compared to the gold standard were 0.8475 and 0.7355, respectively, whereas these measurements for Freesurfer were 0.741 and 0.5915. Our ICC results were also superior over Freesurfer. Additionally, the Bland–Altman plot demonstrated that our method did not show obvious overestimation or underestimation in the hippocampus segmentation. Our approach achieved classification accuracies of 91% for NC vs AD and 75% for NC vs MCI which is close to the HarP classification results.
A clinical decision support system using multi-modality imaging data for disease diagnosis
Published in IISE Transactions on Healthcare Systems Engineering, 2018
Nathan Gaw, Todd J. Schwedt, Catherine D. Chong, Teresa Wu, Jing Li
Structural MRI data were obtained from two Siemens 3 T MRI machines. Using a cortical reconstruction and segmentation program in the FreeSurfer image analysis suite (version 5.3, http://www.surfer.nmr.mgh.harvard.edu/), cortical area, thickness and volume measurements of 68 ROIs were extracted. Additionally, resting-state functional connectivities—i.e., fMRI data—were collected for each subject. Standard Statistical Parametric Mapping (SPM) methods were used to preprocess the fMRI data. Specifically, fMRI signals were temporally filtered between 0.01 to 0.1 Hz to retain the low-frequency components. Variance relating to signals of no interest was removed through linear regression. Thirty-three ROIs were chosen based on commonly cited regions for which PMs show abnormalities (Mainero et al., 2011; Russo et al., 2012). Among the 33 ROIs, there are 16 pairs; each pair consists of two regions with the same name but located at the left and right sides of the brain, respectively. The remaining one ROI is located in the middle of the brain. We aggregated each pair of ROIs into one ROI by averaging their respective time courses. This reduces the number of ROIs to 16+1 = 17. Partial correlations between the 17 ROIs were computed, forming 136 connectivity features. Note that we also tried keeping the original 33 ROIs without pair-wise aggregation, but the result was not as good as the one with aggregation.