Method of analyzing multi-sequence MRI data for analysing brain abnormalities in a subject
Inventors
Crainiceanu, Ciprian • Goldsmith, Arthur Jeffrey • Pham, Dzung • Reich, Daniel S. • Shiee, Navid • Shinohara, Russell T. • Sweeney, Elizabeth M.
Assignees
Johns Hopkins University • Henry M Jackson Foundation for Advancedment of Military Medicine Inc • US Department of Health and Human Services
Publication Number
US-9888876-B2
Publication Date
2018-02-13
Expiration Date
2033-03-21
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
The present invention, referred to as Oasis is Automated Statistical Inference for Segmentation (OASIS), is a fully automated and robust statistical method for cross-sectional MS lesion segmentation. Using intensity information from multiple modalities of MRI, a logistic regression model assigns voxel-level probabilities of lesion presence. The OASIS model produces interpretable results in the form of regression coefficients that can be applied to imaging studies quickly and easily. OASIS uses intensity-normalized brain MRI volumes, enabling the model to be robust to changes in scanner and acquisition sequence. OASIS also adjusts for intensity inhomogeneities that preprocessing bias field correction procedures do not remove, using BLUR volumes. This allows for more accurate segmentation of brain areas that are highly distorted by inhomogeneities, such as the cerebellum. One of the most practical properties of OASIS is that the method is fully transparent, easy to implement, and simple to modify for new data sets.
Core Innovation
The present invention, known as OASIS (Automated Statistical Inference for Segmentation), is a fully automated and robust statistical method designed for cross-sectional multiple sclerosis (MS) lesion segmentation in the brain. It uses intensity information from multiple MRI modalities including T1-weighted, T2-weighted, FLAIR, and PD-weighted images. A logistic regression model assigns voxel-level probabilities indicating the presence of lesions. This model produces interpretable regression coefficients that enable quick and easy application to imaging studies, and the method includes intensity normalization to ensure robustness to changes in scanner and acquisition sequence.
A significant problem addressed by this invention is the variability and complexity of MRI data, which arise from differences in scanner hardware, acquisition parameters, and intensity inhomogeneities that standard bias field corrections fail to remove. These variations affect the accuracy and generalizability of automated lesion segmentation methods. The invention overcomes these challenges by incorporating a sequence of Gaussian blurs (BLUR volumes) that capture residual intensity inhomogeneities and regional features while removing local voxel-specific noise, allowing a multi-scale analysis that improves segmentation accuracy even in regions like the cerebellum which are highly distorted by inhomogeneities.
The method involves constructing initial brain tissue masks excluding cerebrospinal fluid, normalizing MRI volumes, creating multi-resolution BLUR volumes, and fitting logistic regression models iteratively to refine lesion segmentation. The approach isolates probable abnormalities from blurred images, removes these from the data before reblurring, and refits the model for refined voxel-level probability estimates. The fully transparent, easy-to-implement, and adaptable nature of OASIS makes it suitable for different data sets and imaging centers.
Claims Coverage
The claims cover three independent aspects: a method for analyzing multi-sequence MRI data, a system for analysis of such data, and a refined method involving brain tissue mask creation and logistic regression modeling.
Automated statistical logistic regression model for lesion probability assignment
A method that obtains intensity information from multiple MRI modalities and assigns voxel-level probabilities of neurological abnormalities using a logistic regression model that produces regression coefficients applicable to multi-sequence MRI data.
Multi-scale image blurring and intensity inhomogeneity adjustment
Sequential Gaussian blurring of multi-sequence MRI images with varying kernel sizes to oversmooth images while preserving regional features and spatial inhomogeneities, removing local voxel-specific features, and using these blurred images in the logistic regression model for refined lesion detection.
Iterative lesion isolation and reblurring for refined modeling
Isolating probable abnormalities in each modality's blurred images, removing these from the image, reblurring to provide refined regional patterns including spatial inhomogeneity, and refitting the logistic regression model with interactions between normalized and blurred images to produce refined voxel-specific probabilities and binary segmentation maps.
System for multi-sequence MRI data acquisition, storage, and processing
A system comprising an MRI imaging system, data storage, and a computing device configured to process multi-sequence MRI data, apply the logistic regression model with the described blurring and adjustment steps, and generate segmentation identifying neurological abnormalities and their location.
Brain tissue masking and logistic regression model refinement method
A method comprising creating a brain tissue mask, normalizing MRI volumes, constructing initial BLUR volumes, fitting an initial logistic regression model, performing lesion segmentation, reblurring volumes with lesions removed, refitting the logistic regression model, and generating a probability map and lesion segmentation based on refined voxel-specific probabilities.
The inventive claims collectively cover an integrated method and system employing multi-modal MRI intensity normalization, multi-scale Gaussian blurring to correct intensity inhomogeneities, iterative logistic regression modeling with interaction terms to identify neurological abnormalities such as MS lesions, and a practical computing framework to generate interpretable voxel-level probability maps and binary segmentations.
Stated Advantages
Fully automated and robust method for MS lesion segmentation that reduces reliance on manual delineation, saving time and reducing variability.
Robustness to changes in scanner hardware and imaging acquisition parameters due to intensity normalization and use of BLUR volumes to adjust for residual intensity inhomogeneities.
Produces interpretable regression coefficients allowing straightforward application and adaptation to new data sets.
Computationally efficient with segmentation time per study reduced to approximately 30 minutes on a standard workstation, further decreased with parallel computing.
High accuracy and generalizability demonstrated across datasets from different imaging centers with varying acquisition protocols.
Documented Applications
Analysis and segmentation of multiple sclerosis lesions in brain MRI studies using cross-sectional multi-sequence MRI data.
Quantitative evaluation of lesion load and lesion location in subjects with neurological disorders.
Clinical trials involving disease-modifying therapies where MRI lesion quantification serves as an endpoint.
Interested in licensing this patent?