Standardizing Breast Density Assessments Using AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The continuous changes in the BI-RADS standard for breast density classification in mammography lead to inconsistencies in radiologist reports, making it difficult to assess breast cancer risk based on medical history and evaluate changes in breast density over time, as reports may not follow standardized descriptions.
Innovation Solution
A method using a convolutional neural network to standardize breast density classifications by assigning BI-RADS fifth edition labels to images labeled under different editions, and a generative adversarial network to predict mammogram image appearances for current examinations based on prior images, comparing predicted and actual scores to determine if changes are due to subjective assessments or actual density changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the BI-RADS standard is continuously updated with new editions, then the standardization of breast density classification is improved, but inconsistencies in radiologist reports increase due to different editions being used
Solution Approach 1:
The patent creates a standardized reference framework by copying and applying the BI-RADS fifth edition classification criteria to reclassify historical mammogram images from previous editions. This allows consistent evaluation of breast density across different time periods and editions, transforming inconsistent historical data into a uniform format that can be reliably compared with current assessments
Solution Approach 2:
The patent changes the classification parameters by transitioning from percentile-based glandular tissue measurement (4th edition) to purely subjective categorical descriptors (5th edition). This parameter change resolves the contradiction by establishing a new standardized framework that eliminates the need for complex measurements while maintaining reliability through consistent categorical classification across all reports
2Adaptability or versatility
If different BI-RADS editions are used for mammogram interpretation, then radiologists can use the most current standard available, but it becomes difficult to assess patient risk based on medical history
Solution Approach 1:
The patent introduces a standardized reference framework as an intermediary layer between different BI-RADS editions. This framework uses the fifth edition criteria as a common language to translate and compare breast density classifications from various editions, enabling accurate longitudinal assessment of patient risk without requiring re-interpretation of historical images under different standards
Solution Approach 2:
The patent performs preliminary reclassification of historical mammogram images using the fifth edition standard before conducting risk assessment. This preliminary action ensures that all historical data is transformed into the current standardized format, allowing radiologists to directly compare past and present breast density without needing to recall or reference multiple different classification systems
3Ease of operation
If subjective breast density descriptors are used without standardized criteria, then radiologist reporting flexibility is improved, but evaluation of breast density changes over time becomes difficult
Solution Approach 1:
The patent makes the fifth edition categorical descriptors universally applicable by establishing them as the standardized reference framework for all breast density assessments. These four categories (almost entirely fatty, scattered areas of fibroglandular density, heterogeneously dense, and extremely dense) serve multiple functions: they maintain radiologist reporting flexibility while simultaneously providing precise, consistent measurement criteria for evaluating breast density changes over time
Data Source
AI summary
A method, system and computer program product for determining changes in breast density. A generative adversarial network is trained to predict an appearance of a mammogram image for a patient's current examination based on mammogram images assigned labels of a first type of density classification. An appearance of a mammogram image for a patient's current examination is predicted using the generative adversarial network based on a mammogram image(s) obtained from the patient's prior examination assigned labels of the first type of density classification. A comparison is made between the predicted and actual mammogram images for the patient's current examination to determine if there is a difference between scores assigned to the predicted and actual mammogram images, and if so, whether such difference can be attributed to the subjective assessment by different physicians or changes in the standards of density classification or is due to a real change in breast density.