Automated cancer detection using MRI
Inventors
Kwak, Jin Tae • Wood, Bradford J. • Xu, Sheng • Turkbey, Baris • Choyke, Peter L. • Pinto, Peter A. • Summers, Ronald M.
Assignees
US Department of Health and Human Services
Publication Number
US-10215830-B2
Publication Date
2019-02-26
Expiration Date
2035-12-16
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Methods and systems for diagnosing cancer in the prostate and other organs are disclosed. Exemplary methods comprises extracting texture information from MRI imaging data for a target organ, sometimes using two or more different imaging modalities. Texture features are determined that are indicative of cancer by identifying frequent texture patterns. A classification model is generated based on the determined texture features that are indicative of cancer, and diagnostic cancer prediction information for the target organ is then generated to help diagnose cancer in the organ.
Core Innovation
The disclosed invention relates to methods and systems for diagnosing cancer in prostate and other organs using imaging data, particularly magnetic resonance imaging (MRI). The invention extracts texture information from MRI imaging data of a target organ and determines texture features indicative of cancer by identifying frequent texture patterns. A classification model is generated based on these discriminative texture features, allowing the system to generate diagnostic cancer prediction information or a cancer prediction map for the whole organ to aid cancer diagnosis.
The problem addressed by the invention arises from conventional prostate cancer diagnosis relying on random biopsy guided by ultrasound imaging, which leads to overdiagnosis of incidental tumors and underdiagnosis of clinically significant lesions outside typical biopsy templates. Similar shortcomings exist in diagnosing other cancers such as brain, breast, colon, gallbladder, liver, and pancreas cancers. While multiparametric MRI can visualize aggressive lesions and improve detection, the examination process is complex and time consuming, requiring expert knowledge to integrate multiple MRI sequences and conflicting imaging information.
To overcome these limitations, the invention employs computer-aided diagnosis (CAD) systems to assist in processing multiparametric MRI by extracting and highlighting meaningful texture information. The systems can utilize various MRI sequences such as T2-weighted MRI and high-b-value diffusion weighted imaging (DWI), among others, and apply texture operators (local binary pattern, local direction derivative pattern, variance measure) to extract robust texture features. A multi-stage feature selection method including frequent pattern mining, Wilcoxon rank-sum testing, and minimum redundancy maximum relevance (mRMR) criterion identifies the most discriminative features for cancer detection. These features are then used to build a support vector machine classification model, which outputs cancer prediction maps for diagnosis.
Claims Coverage
The claims disclose two independent claims defining a method and a computing system for cancer diagnosis using texture feature analysis of imaging data. The coverage includes a detailed multi-stage process of discovering, selecting, and utilizing discriminative texture patterns for cancer prediction.
Texture feature extraction using frequent pattern mining
A process of extracting texture information from imaging data and identifying frequent texture patterns by mining length-1 frequent patterns and generating longer frequent patterns through analysis of sub-datasets, to determine texture features indicative of cancer.
Selection of significant texture patterns using statistical and relevance criteria
Selection of significant texture features by comparing frequent patterns between cancer and benign tissue using Wilcoxon rank-sum test, followed by ordering via minimum redundancy maximum relevance (mRMR) criterion and choosing the most discriminative features through forward feature selection.
Generation of classification model and diagnostic output
Building a classification model based on the selected discriminative texture features using support vector machine (SVM), and generating diagnostic cancer prediction information or a cancer prediction map for the target organ.
Use of multimodal imaging data and preprocessing
Use of imaging data derived from different modalities such as high-b-value diffusion weighted MRI and T2-weighted MRI, with optional normalization and registration of imaging datasets prior to texture analysis.
Together, the claims cover a computer-aided diagnosis approach that applies advanced texture pattern mining and feature selection methods on multimodal MRI data to generate a cancer classification model, providing diagnostic prediction to facilitate cancer detection.
Stated Advantages
Significantly improves detection rate of clinically significant cancers by identifying discriminative texture features in multiparametric MRI.
Reduces negative biopsies by distinguishing cancer from MR-positive benign lesions preselected by expert radiologists.
Facilitates biopsy and improves diagnostic yield, workflow, and throughput of prostate biopsy procedures.
Assists nonspecialists in interpreting prostate and other organ MRI, potentially reducing the learning curve for interpretation training.
Enables whole-organ cancer prediction maps that correspond well with biopsy-proven cancer lesions, aiding targeted diagnosis and treatment planning.
Documented Applications
Detection and diagnosis of prostate cancer using multiparametric MRI datasets including T2-weighted MRI and high-b-value diffusion weighted MRI.
Application to cancer detection in other organs including brain, breast, colon, gallbladder, liver, kidneys, lungs, bones, and pancreas.
Generating cancer prediction maps for whole prostate or other organs to aid in biopsy targeting and treatment planning.
Facilitating computer-aided diagnosis in clinical settings by assisting interpretation of multiparametric imaging for improved workflow and diagnostic accuracy.
Interested in licensing this patent?