Machine Learning Mammography Exposure Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic exposure control (AEC) techniques in mammography often fail to accurately identify the densest breast regions, leading to either excessive or inadequate radiation exposure during mammographic imaging, resulting in suboptimal image quality.
Innovation Solution
The use of machine learning (ML) models trained on datasets containing patient and imaging data to predict optimal exposure technique parameters in real-time, eliminating the need for traditional sampling methods and scout images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional AEC sampling methods are used to identify densest breast regions, then the process is simple and quick, but the identification accuracy is insufficient leading to overexposure or underexposure
Solution Approach 1:
The patent replaces the traditional mechanical sampling method with a machine learning-based system that uses neural networks to analyze mammographic images and predict optimal exposure techniques. The ML model processes image data to identify densest breast regions more accurately than manual sampling, resolving the contradiction between measurement precision and system complexity by using intelligent algorithms rather than physical sampling mechanisms
Solution Approach 2:
The system changes the parameters used for exposure control by predicting optimal kVp and mAs values based on patient-specific attributes and image characteristics. Instead of using fixed sampling thresholds, the system dynamically adjusts exposure parameters based on learned patterns from training data, improving identification accuracy while maintaining computational efficiency
2Manufacturing precision
If exposure technique parameters are optimized for each patient, then image quality improves, but the time required for processing increases
Solution Approach 1:
The system performs preliminary action by pre-training the neural network model on a comprehensive dataset of patient attributes and imaging parameters before actual use. This pre-training allows the model to make rapid predictions during patient imaging without requiring time-consuming analysis, as the complex pattern recognition has already been performed during the training phase
Solution Approach 2:
The patent uses copying by creating a digital model (neural network) that replicates the complex relationships between patient attributes, imaging parameters, and optimal exposure techniques. This digital copy allows for rapid inference and prediction without repeatedly performing complex calculations, thus maintaining high image quality consistency while reducing processing time
3Reliability
If scout images are used to determine exposure techniques, then exposure optimization is possible, but radiation exposure to patient increases and imaging time extends
Solution Approach 1:
The patent extracts the essential information needed for exposure optimization directly from the mammographic image data and patient attributes, eliminating the need for separate scout images. The neural network model processes the actual diagnostic image data to predict optimal exposure parameters, thereby removing the harmful scout image step while maintaining reliable exposure optimization
Solution Approach 2:
The system converts the potentially harmful effect of increased radiation exposure into a benefit by using machine learning to predict optimal exposure parameters that minimize radiation while maintaining image quality. The ML model learns from training data to identify the minimum effective exposure parameters for each patient, turning what would normally be excessive radiation into optimized, minimal necessary radiation
Data Source
AI summary
Examples of the present disclosure describe systems and methods for using machine learning (ML) to predict optimal exposure technique for acquiring mammographic images. In aspects, training data may be collected from one or more data sources. The training data may comprise sample data of patient attribute data and sample image attribute data, image metadata, image pixel data, and/or exposure technique parameters. The training data may be used to train an ML model to determine the optimal exposure technique parameters for acquiring mammographic images of a patient. After the ML model has been trained, patient data that is collected from a patient during a patient visit may be provided as input to the ML model. The ML model may output optimal exposure technique parameters for the patient in real-time. The real-time output of the ML model may be used to generate one or more mammographic images of the patient.