AI-Guided Mammography Positioning With On-System Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mammography systems face challenges in accurately positioning the breast prior to imaging, leading to radiation dose duplication and increased recall rates due to the need for additional imaging after improper positioning, and the training of artificial intelligence for breast positioning requires significant data collection and regulatory compliance.
Innovation Solution
An x-ray mammography system with on-site sensors and a training module that utilizes continuous data from normal operations to train an AI model for breast and patient positioning, providing real-time feedback without altering the clinical workflow or transmitting patient data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If camera-based AI model is used to evaluate breast positioning in real-time, then positioning accuracy is improved, but training data collection requirements increase system complexity
Solution Approach 1:
The system uses itself to generate training data by capturing camera images and corresponding x-ray images during normal operation, eliminating the need for external data collection systems. The AI model is trained using internally generated data pairs, making the system self-sufficient for training purposes.
Solution Approach 2:
The system continuously collects training data during normal mammography operations without interrupting workflow. Camera images and x-ray images are captured simultaneously throughout the procedure, allowing ongoing accumulation of training data while maintaining continuous useful action.
2Measurement precision
If large training dataset is collected from external sources, then AI model accuracy is improved, but data privacy compliance time and expense increase
Solution Approach 1:
The system generates its own training data internally during normal operation, eliminating the need to collect data from external sources and avoiding associated privacy compliance requirements. This self-service approach to data generation resolves the time and expense burden of external data collection.
Solution Approach 2:
The system creates copies of actual mammography data (camera images paired with x-ray images) for training purposes without transmitting or storing sensitive patient information externally. The training dataset is a local copy generated and retained within the system, maintaining privacy while enabling model training.
3Measurement precision
If breast positioning is evaluated after x-ray image acquisition, then diagnostic accuracy is improved, but radiation dose is increased due to duplication
Solution Approach 1:
The system performs preliminary evaluation of breast positioning using camera images before acquiring the x-ray image. By assessing positioning accuracy in advance, the system prevents unnecessary radiation exposure that would result from re-imaging improperly positioned breasts.
Solution Approach 2:
The system provides real-time feedback on breast positioning quality before x-ray image acquisition. This feedback loop allows technologists to correct positioning issues prior to imaging, eliminating the need for duplicate imaging and reducing overall radiation dose to the patient.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various methods and systems are provided for breast positioning assistance during mammography and image guided interventional procedures. In one example, a sensor detection system (100,252) is utilized to evaluate one or more of a patient position, a breast position, and breast anatomy to determine if the patient (236) and breast (236) are adjusted to desired positions preferred for a desired view and imaging procedure prior to acquiring x-ray images, with real-time feedback provided to guide the user to position the breast and/or the patient based on the evaluation. The evaluation is performed by an artificial intelligence (Al) model (262) stored on the x-ray mammography system (10,210 that is trained using sensor data (306) obtained from imaging procedures performed using the x-ray mammography system (10,210) and fashioned into training datasets (320) by a training module (263) located on the x-ray mammography system (10,210). The datasets (320) are supplied to the AI model (262) without any transmission of the datasets (320) exteriorly of the mammography system (10,210).