Classification based on characterization analysis methods and systems
Inventors
Jaber, Mustafa • Beziaeva, Liudmila A • Szeto, Christopher W • Song, Bing
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11676707-B2
Publication Date
2023-06-13
Expiration Date
Abstract
A method at a computing device for classifying elements within an input, the method including breaking the input into a plurality of patches; for each patch: creating a vector output; applying a characterization map to select a classification bin from a plurality of classification bins; and utilizing the selected classification bin to classify the vector output to create a classified output; and compiling the classified output from each patch.
Core Innovation
The invention classifies elements within image data of a histopathology slide by breaking the image data into a plurality of patches, where each patch corresponds to a portion of the image data. For each patch, the system creates a vector output and applies a density map to obtain a cell density. The method then selects a classification bin from a plurality of classification bins according to the cell density.
Each classification bin corresponds to a different density range for the element, and the selected classification bin is utilized to classify the vector output to create a classified output. After classifying each patch, the method compiles the classified output from each patch into an image-level result. The approach preserves spatial resolution while generating patch-level outputs and combining them into a final result.
In embodiments, the invention uses tumor masking, including expert-guided and deep-learning generated tumor mask options, before generating the vector output. Alternative characterization maps are also described, while the characterization map is typically a cell-density map. The described implementations include density-stacked bin classifiers that use both linear and non-linear classifiers and are compared against baseline single-classifier approaches using ROC curve, AUC, and accuracy.
Example applications include lung non-small-cell cancer classification for adenocarcinoma versus squamous cell carcinoma and breast cancer subtyping for Luminal A, Luminal B, triple-negative/basal-like, and HER2-enriched categories. The described cell-density related characterization is linked to selecting the appropriate classification bin, and the resulting patch-level classifications are compiled to support the overall histopathology slide classification outcomes.
Claims Coverage
The document includes three independent claims covering a method, a computing device, and a non-transitory computer readable medium. Each independent claim is built around one core pipeline with additional structural refinements identified in dependent claims, including density map-driven bin selection and bin-based classification of patch vector outputs.
Density-map bin selection for patch vector classification
Breaking the image data into a plurality of patches corresponding to portions of the image data, creating a vector output for each patch, applying a density map to obtain a cell density, selecting a classification bin from a plurality of classification bins according to the cell density where each bin corresponds to a different density range, utilizing the selected classification bin to classify the vector output to create a classified output, and compiling the classified output from each patch.
Computing device for density-map bin patch classification
A computing device configured to break the image data into a plurality of patches, create a vector output for each patch, apply a density map to obtain a cell density, select a classification bin according to the cell density with bins corresponding to different density ranges, utilize the selected classification bin to classify the vector output to create a classified output, and compile the classified output from each patch.
Non-transitory computer readable medium for density-map bin patch classification
A non-transitory computer readable medium storing instruction code that, when executed, breaks the image data into a plurality of patches, creates a vector output for each patch, applies a density map to obtain a cell density, selects a classification bin from a plurality of classification bins according to the cell density where each bin corresponds to a different density range, utilizes the selected classification bin to classify the vector output to create a classified output, and compiles the classified output from each patch.
Across the independent claims, the main inventive coverage centers on combining patch-wise vector outputs with density-map-based selection of density-range classification bins, followed by compiling patch classified outputs into a final result.
Stated Advantages
Improved performance for density-stacked bin classifiers versus baseline single-classifier approaches.
Preserves spatial resolution while generating and compiling patch-level outputs.
Documented Applications
Lung non-small-cell cancer classification of adenocarcinoma versus squamous cell carcinoma using patch-level density-based classification compiled into an image-level result.
Breast cancer subtyping into Luminal A, Luminal B, triple-negative/basal-like, and HER2-enriched categories using density-based classification bins compiled into an image-level result.
Interested in licensing this patent?