Neural network classification of osteolysis and synovitis near metal implants
Inventors
Koch, Kevin M. • NENCKA, Andrew S. • KARR, Robin A. • Swearingen, Bradley J. • Potter, Hollis • Koff, Matthew F.
Assignees
Medical College of Wisconsin • New York Society for Relief of Ruptured and Crippled
Publication Number
US-11969265-B2
Publication Date
2024-04-30
Expiration Date
2039-03-04
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Systems and methods for training and implementing a machine learning algorithm to generate feature maps depicting spatial patterns of features associated with osteolysis, synovitis, or both. MRI data, including multispectral imaging data, are input to the trained machine learning algorithm to generate the feature maps, which may indicate features such as a location and probability of a pathology classification, a severity of synovitis, a type of synovitis, a synovial membrane thickness, and other features associated with osteolysis or synovitis. In some implementations, synovial anatomy are segmented in the MRI data before inputting the MRI data to the machine learning algorithm. These segmented MRI data may be generated using another trained machine learning algorithm.
Core Innovation
The invention provides systems and methods for training and implementing a machine learning algorithm to generate feature maps that depict spatial patterns or distributions of features associated with osteolysis, synovitis, or both. Magnetic resonance imaging (MRI) data, which may include multispectral imaging data, are input into the trained machine learning algorithm to produce these feature maps. The generated feature maps may indicate important features such as the location and probability of a pathology classification, the severity of synovitis, the type of synovitis, the synovial membrane thickness, and other features relevant to osteolysis and synovitis.
A key part of the invention is the use of neural network-based machine learning algorithms, particularly convolutional neural networks and encoder-decoder architectures, that are trained on labeled MRI datasets. These machine learning models are built to classify and localize pathological features near metal implants, and can generate visual outputs that assist clinicians in identifying disease manifestations. In some implementations, synovial anatomy is segmented in the MRI data—potentially by another trained machine learning algorithm—before being input to the classification model.
The problem addressed by the invention stems from the complexity of interpreting MRI data near hip arthroplasty (HA) and other metallic implants, where a variety of abnormal tissue signatures are visible but require advanced expertise and may be underutilized in non-specialist clinical environments. Previous advances, such as 3D multi-spectral imaging (3D-MSI), have mitigated metal artifacts but have also exposed the difficulty of interpreting the nuanced and complex MRI findings near implants. This results in poor characterization of clinically relevant MRI signatures and an under-utilization of MRI technology for orthopedic imaging.
Claims Coverage
There are three independent claims covering methods for generating feature maps for synovitis classification, constructing machine learning algorithms for generating synovitis feature maps, and generating synovial membrane thickness data.
Generating feature maps for synovitis classification using trained machine learning algorithms
A method comprising: - Accessing MRI data acquired from a subject with a computer system. - Accessing a trained machine learning algorithm that has been trained on data to generate feature maps associated with synovitis. - Inputting the MRI data into the trained machine learning algorithm to generate output, including at least one feature map depicting the spatial distribution of a feature associated with synovitis. The feature map specifically depicts the spatial distribution of classifications indicating types of synovitis at different locations.
Constructing and implementing a machine learning algorithm to generate spatial feature maps of synovitis patterns
A method comprising: - Constructing a trained machine learning algorithm by accessing training data that includes MRI data from multiple subjects and labeled data indicating features associated with synovitis. - Training the machine learning algorithm to generate a feature map indicating spatial patterns of those features across a region-of-interest. - Generating a feature map by inputting MRI data from a subject into the trained algorithm, where the feature mapped is a type of synovial reaction causing synovitis.
Generating feature data indicating synovial membrane thickness using a machine learning algorithm
A method comprising: - Accessing MRI data from a subject with a computer system. - Accessing a trained machine learning algorithm that has been trained to generate feature data indicative of synovial membrane thickness. - Inputting the MRI data into the algorithm to generate output that indicates the thickness of the synovial membrane in the subject.
The independent claims define inventive methods for: (1) generating detailed synovitis classification maps from MRI using machine learning; (2) constructing and using machine learning models, particularly neural networks, for mapping spatial synovitis patterns; and (3) producing quantitative synovial thickness data from MRI using trained algorithms.
Stated Advantages
The systems and methods provide supplementary guidance for radiological identification of soft-tissue pathologies near hip arthroplasty or metallic implants.
Feature maps generated by the invention allow for direct visual presentation of regional tissue abnormalities to radiologists and surgeons.
The invention adds simplicity and intuition to clinical MRI application workflows, accelerating and broadening the adoption of advanced MRI utility in hip arthroplasty management.
The approach can aid education and comprehension of tissue signatures near implants in large patient cohorts, supporting generation of natural history datasets.
The invention enables quantitative evidence to inform severity of tissue disease in failing joint replacements and supports evaluation of revision procedures.
Machine learning-based soft tissue pathology differentiation provides robust monitoring and tracking of synovitis characteristics, including severity and type.
Documented Applications
Use in computer-aided diagnosis to provide radiological guidance for identifying osteolysis and synovitis near hip arthroplasty or other metallic implants.
Educational tool for radiologists and clinicians to understand and interpret MRI features near joint replacements or metallic orthopedic implants.
Facilitating and broadening the clinical adoption of advanced MRI imaging and analysis in orthopedic imaging workflows, particularly in hip arthroplasty management.
Enabling quantitative assessment and monitoring of tissue disease severity to inform the necessity and planning of surgical revision procedures for joint replacements.
Interested in licensing this patent?