Automated Mammogram Segmentation for Reproducible Density Measurement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining mammographic density are subjective and lack reproducibility, leading to unacceptably high intra- and inter-observer variability, which hampers accurate and reproducible breast cancer risk prediction.
Innovation Solution
A computing system uses a machine learning model, specifically a U-net architecture, to automatically segment mammograms and determine fibroglandular density, providing a quantitative and reproducible assessment of breast cancer risk by generating segmentation maps and classifying subjects into risk levels based on density values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If subjective assessment by radiologists using BI-RADS is used, then clinical practice is simplified, but measurement precision and reliability deteriorate due to high intra- and inter-observer variability
Solution Approach 1:
The patent replaces the manual, subjective visual assessment mechanism (radiologist observation and categorization) with an automated computational image analysis system using machine learning models. This substitution eliminates human variability while maintaining ease of use through automated processing of mammographic images to generate quantitative density measurements.
Solution Approach 2:
The system enables self-service by allowing the mammographic density assessment to be performed automatically without requiring radiologist intervention for the actual measurement. The ML model independently processes images, segments fibroglandular tissue, calculates density values, and generates risk assessments, making the measurement process autonomous and reproducible.
2Measurement precision
If automated machine learning methods are implemented, then measurement precision and reliability improve, but device complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the complex task of breast density assessment into distinct computational stages: image preprocessing, fibroglandular tissue segmentation using U-net architecture, density calculation from segmented regions, and risk categorization. This modular approach manages system complexity by breaking down the ML pipeline into separable, well-defined components that can be developed and validated independently.
Solution Approach 2:
The patent introduces segmentation maps as an intermediary representation between the input mammographic image and the final density measurement. These maps serve as a bridge that visually and computationally delineate fibroglandular tissue regions, making the complex ML processing transparent and verifiable while maintaining measurement precision.
3Reliability
If quantitative automated measurement is implemented, then reliability and reproducibility improve, but loss of information may occur through automated processing
Solution Approach 1:
The patent implements feedback by generating segmentation maps that visually display the identified fibroglandular tissue regions overlaid on or alongside the original mammographic image. This feedback mechanism allows clinicians to verify that the automated segmentation accurately represents the actual tissue structures, ensuring that no critical diagnostic information is lost while maintaining quantitative reproducibility.
Data Source
AI summary
Presented herein are systems and methods of determining density values from mammograms. A computing system may identify a first mammogram of a first breast region of a first subject. The first mammogram may have a first region of interest (ROI) corresponding to a first dense area of the first breast region. The computing system may apply the first mammogram to a machine learning (ML) model to generate a first segmentation map identifying the first ROI within the first mammogram. The computing system may determine a density value for the first dense area of the first breast region based on the first segmentation map.