System and method of automatically detecting tissue abnormalities
Inventors
Crainiceanu, Ciprian M. • Sweeney, Elizabeth M. • Shinohara, Russell T. • Goldsmith, Arthur J. • Reich, Daniel • Shea, Colin
Assignees
Johns Hopkins University • US Department of Health and Human Services
Publication Number
US-9607392-B2
Publication Date
2017-03-28
Expiration Date
2032-12-05
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A method of automatically detecting tissue abnormalities in images of a region of interest of a subject includes obtaining first image data for the region of interest of the subject, normalizing the first image data based on statistical parameters derived from at least a portion of the first image data to provide first normalized image data, obtaining second image data for the region of interest of the subject, normalizing the second image data based on statistical parameters derived from at least a portion of the second image data to provide second normalized image data, processing the first and second normalized image data to provide resultant image data, and generating a probability map for the region of interest based on the resultant image data and a predefined statistical model. The probability map indicates the probability of at least a portion of an abnormality being present at locations within the region of interest.
Core Innovation
The invention provides a method, system, and computer-readable medium for automatically detecting tissue abnormalities in images of a region of interest of a subject. This involves obtaining first and second image data for the region of interest, normalizing these data based on statistical parameters derived from their respective images to produce normalized image data, processing these normalized data to produce resultant image data, and generating a probability map of the region. This probability map indicates the likelihood that abnormalities are present at specific locations within the region of interest.
The specific problem addressed is the difficulty and inefficiency of manually detecting and tracking lesions, such as those in multiple sclerosis, in sequential imaging studies. Manual segmentation is time-consuming, costly, and prone to human error, particularly given that lesions represent a small proportion of all tissue changes. Prior art methods involving subtraction images from MRI modalities face artifacts from misregistration and partial volume effects, and have not effectively combined multiple imaging modalities. Thus, there is a need for improved, automated systems and methods that can incorporate information from multiple imaging modalities for sensitive and specific detection of tissue abnormalities over time.
Claims Coverage
The patent includes three independent claims directed respectively to a method, a computer-readable medium, and a system for automatically detecting tissue abnormalities. Each claim features processes involving normalization of segmented sub-images, processing of normalized image data, and generation of a probability map based on a predefined statistical model.
Normalization of segmented image data based on sub-image statistical parameters
The method, medium, and system obtain image data for a region of interest, segment the images into multiple sub-images corresponding to anatomical structures, and normalize each sub-image individually using statistical parameters derived from its image elements, rather than normalizing the entire image globally.
Processing of normalized image data from multiple time points
The inventive approach processes normalized normalized data obtained from at least two different image data sets of the same region of interest, which may be collected at different times or from different modalities, to provide resultant image data for further analysis.
Generation of a probability map indicating tissue abnormality presence using a predefined statistical model
Based on the resultant image data, a probability map is generated that indicates the probability of at least a portion of an abnormality at specific locations within the region of interest, wherein the statistical model employed may be a logistic regression model or others predefined.
These inventive features together provide a comprehensive approach to automatically detect tissue abnormalities by utilizing normalized sub-image data, processing multiple time-point data sets, and generating statistical probability maps, applicable in various imaging modalities and configurations.
Stated Advantages
The method provides an automatic and computationally fast solution for detecting tissue abnormalities, reducing the time, cost, and variability associated with manual segmentation.
It incorporates information from multiple imaging modalities, improving robustness against artifacts and increasing sensitivity and specificity in lesion detection.
The method is adaptable to different imaging modalities, varying scanning parameters, and magnetic field strengths, making it broadly applicable.
The solution is artifact-resistant by using longitudinal data, smoothing predictions, and leveraging multiple modalities, enhancing accuracy in real-world scenarios.
Documented Applications
Detection and longitudinal tracking of new and enlarging brain lesions in multiple sclerosis patients using MRI modalities such as FLAIR, PD, T2-weighted, and T1-weighted images.
Replacement of manual lesion segmentation in clinical trials to assess lesion volume changes as markers of disease progression and response to therapy in multiple sclerosis.
General applications for monitoring changes and pathologies in co-registered serial images across modalities, including volumetric changes in vascular disease, tumors, and lesion repair and shrinkage.
Potential application to other organs such as lung, liver, and kidneys, and other imaging techniques including positron emission tomography-computed tomography studies in oncology.
Interested in licensing this patent?