Title: Magnetic Resonance Connectome Automated Pipeline
ArXiv ID: 1111.2660
Date: 2023-06-15
Authors: : John Smith, Jane Doe, Robert Johnson
📝 Abstract
This manuscript presents a novel, tightly integrated pipeline for estimating a connectome, which is a comprehensive description of the neural circuits in the brain. The pipeline utilizes magnetic resonance imaging (MRI) data to produce a high-level estimate of the structural connectivity in the human brain. The Magnetic Resonance Connectome Automated Pipeline (MRCAP) is efficient and its modular construction allows researchers to modify algorithms to meet their specific requirements. The pipeline has been validated and over 200 connectomes have been processed and analyzed to date. This tool enables the prediction and assessment of various cognitive covariates, and this research is applicable to a variety of domains and applications. MRCAP will enable MR connectomes to be rapidly generated to ultimately help spur discoveries about the structure and function of the human brain.
💡 Deep Analysis
Deep Dive into Magnetic Resonance Connectome Automated Pipeline.
This manuscript presents a novel, tightly integrated pipeline for estimating a connectome, which is a comprehensive description of the neural circuits in the brain. The pipeline utilizes magnetic resonance imaging (MRI) data to produce a high-level estimate of the structural connectivity in the human brain. The Magnetic Resonance Connectome Automated Pipeline (MRCAP) is efficient and its modular construction allows researchers to modify algorithms to meet their specific requirements. The pipeline has been validated and over 200 connectomes have been processed and analyzed to date. This tool enables the prediction and assessment of various cognitive covariates, and this research is applicable to a variety of domains and applications. MRCAP will enable MR connectomes to be rapidly generated to ultimately help spur discoveries about the structure and function of the human brain.
📄 Full Content
During the past few years, there have been a number of research groups working on methods and techniques for generating MR connectomes. This complex task requires tools such as FSL/Freesurfer [6], [7], MedINRIA [8], or BrainVISA [9] that combine cortical labeling and segmentation with diffusion tensor imaging. Custom software scripts are typically developed to link the necessary routines and obtain the required connectivity measurements (though see [10]).
In this article we present a new automated method to obtain MR connectomes. Our pipeline generates connectomes through an integrated graphical programming environment, which offers a tightly coupled set of software routines that are available for multiple platforms and input data types. The pipeline is based on the Java Image Science Toolkit (JIST) [11], [12], which runs in conjunction with the Medical Image Processing, Analysis and Visualization (MIPAV) software [13].
Our MR connectome automated pipeline (MRCAP) combines diffusion-weighted images with structural MR images to generate an MR connectome derived from connectivity measurements between anatomically-defined cortical regions.
The connectome is represented by a connectivity matrix suitable for input to graph theoretic [14], [15] or statistical [5], [16][17][18][19][20] algorithms which can infer meaning from the data. The latest stable release of MRCAP is available for download from NITRC at www.nitrc.org/projects/mrcap
, and the code base is actively being developed. Routines (called modules) can be updated and replaced as needed to meet the requirements of the individual researcher. Indeed, because there is no scientific consensus on how to best estimate synaptic connectivity from MRI data, the pipeline is designed to serve as a testbed to explore different approaches.
MRCAP is comprised of three JIST layouts-structural, diffusion, and connectivity-each of which consists of a collection of modules (see Figure 1). The modules themselves A are assembled from a variety of algorithms, authors, and methods; our contribution is the integration of these tools into one automated processing flow. The structural layout operates on the structural MR data and performs skull stripping and parcellation of the brain into labeled gyral regions. The diffusion layout estimates tensors, computes fractional anisotropy (FA) values, and performs fiber tracing. Finally, the connectivity layout registers the diffusion data to the structural space, combines the putative fiber tracts with the associated gyral region labels, and uses the FA information to generate a connectome (in the form of a mathematicallyconvenient adjacency matrix). Alternate measures of connection strength (such as fiber count and fiber length) are also available. A more detailed explanation of the pipeline functionality is described below.
The JIST environment allows a researcher to graphically design a processing pipeline that generates a variety of useful outputs for validation and further analysis. It also provides an Application Programming Interface (API) that facilitates the interoperability of modules developed by various authors. Graphical tools allow for easily adding, swapping, and/or modifying modules as needed. Because it is based on the Java programming language, JIST can run on many computer architectures [12]. Figure 2 shows a screenshot of the JIST programming environment and our pipeline. In the JIST window, the processing steps are represented by input file modules (blue), algorithm modules (red), and lines connecting inputs and outputs (black). Within the JIST framework, researchers can select from a variety of pre-defined modules and assign processing parameters. Both JIST and MIPAV are freely available for download at www.nitrc.org/projects/jist
and www.nitrc.org/projects/mipav
, respectively.
The pipeline accepts the diffusion and structural MR data from a subject, the associated metadata, and user-specified parameters as inputs. Various input image formats are supported through built-in JIST modules, including XML, PAR/REC, NIFTI, and DICOM.
S1: Structural image processing begins with the SPECTRE algorithm [21], which removes the skull and other non-brain tissue using a joint registration and tissue classification technique. The tissue classification is performed using FANTASM, a robust fuzzy C-means intensity classification algorithm [22]. This allows for the identification of highintensity skin and adipose tissue and low-intensity bone matter, all of which can be subsequently eliminated. This
In the second step of the structural processing layout, the brain is divided into a set of 70 regions defined by the Desikan gyral label atlas [23]. Parcellation is achieved by registering one or more template brains to the subject brain using VABRA, a vectorized form of the Adaptive Bases registration Algorithm (ABA) [24]. This algorithm performs nonrigid intensity-based registration using normalized mutual information as a cost
