Mammogram Segmentation for Breast Positioning Quality Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Poor breast positioning in digital mammography can lead to reduced image quality, increased radiation exposure, and difficulty in detecting tumors, especially in patients with breast implants, due to poor breast positioning hiding tumors.
Innovation Solution
A machine learning model is trained to segment digital breast images into key regions of interest, assess image quality, and customize image processing or reconstruction based on segmentation, using a deep convolutional neural network like U-Net to generate masks for breast and other features, and combine them to improve image quality and sensitivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If breast positioning is not optimized, then imaging time and radiation exposure increase due to additional imaging requirements, but image quality and tumor detection sensitivity deteriorate
Solution Approach 1:
The system performs preliminary breast positioning assessment using machine learning analysis of the acquired image before final diagnosis. The ML model evaluates positioning quality metrics (breast coverage, pectoralis muscle visibility, nipple position) immediately after image acquisition, allowing technicians to correct positioning issues before proceeding to diagnostic interpretation or additional imaging, thus preventing time loss from repeat exposures
Solution Approach 2:
The system provides real-time feedback on breast positioning quality through automated ML-based assessment. The model generates quantitative metrics about positioning adequacy (e.g., percentage of breast tissue included, visibility of anatomical landmarks) and compares them against optimal ranges, enabling immediate corrective action to improve image quality without requiring additional imaging exposures
2Reliability
If breast positioning is poor, then additional imaging is required which increases radiation exposure, but this resolves the issue of reduced sensitivity
Solution Approach 1:
The system performs preliminary evaluation of positioning adequacy using machine learning analysis immediately after image acquisition. By assessing breast coverage, pectoralis muscle visibility, and anatomical landmark positioning before diagnostic interpretation, the system identifies positioning deficiencies that would compromise tumor detection sensitivity, allowing corrective action to be taken before additional radiation exposure is administered
Solution Approach 2:
The ML-based system provides quantitative feedback on positioning quality metrics that directly affect tumor detection sensitivity. The model evaluates whether critical regions (breast tissue, pectoralis muscle, nipple) are adequately visualized and provides guidance for repositioning, enabling optimization of detection sensitivity without requiring repeat imaging and associated radiation exposure
3Reliability
If breast positioning is poor in patients with implants, then tumors may be hidden by the implant, but proper positioning techniques are needed to reveal them
Solution Approach 1:
The system performs preliminary analysis of the acquired image using machine learning to assess whether breast positioning adequately displaces implants from potential tumor locations. The ML model evaluates the spatial relationship between implants and breast tissue, identifying cases where positioning may have obscured lesions, allowing radiologists to request targeted re-imaging with adjusted positioning before final diagnosis
Solution Approach 2:
The system provides specialized feedback for implant-associated cases, analyzing whether the implant position obscures breast tissue regions that should be evaluated for tumors. The ML model generates metrics about implant-tissue overlap and visibility of posterior breast structures, enabling technologists to adjust positioning to maximize tumor visibility while minimizing implant interference
Data Source
AI summary
This document describes methods and systems for training a machine learning model to segment digital breast images into key regions of interest, and also for using the model on a new digital breast image to assess whether the breast image exhibits adequate image quality, report the image quality, and/or to use these data to reconstruct or process the new breast image.