2D/3D Mammography Neural Networks for Faster Malignancy Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mammography techniques, including 2D and 3D imaging, suffer from high false positive and false negative rates, non-uniform expert reader availability, and inefficiencies in analyzing breast tissue data, which limits the accuracy and efficiency of cancer detection.
Innovation Solution
A computerized method using a trained neural network to analyze 2D and 3D mammography data, incorporating a backbone portion, regression portion, and classification portion, trained on annotated patches and images to generate malignancy likelihood scores, and outputting indicators for breast tissue analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If digital breast tomosynthesis (DBT) is used to provide three-dimensional imaging, then cancer detection accuracy is improved, but analysis time increases significantly
Solution Approach 1:
The patent segments the 3D breast imaging data into multiple 2D slices and uses a neural network to process each slice independently, then aggregates the results. This segmentation allows the system to maintain 3D imaging accuracy while reducing processing time by leveraging efficient 2D processing architectures.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the raw 3D imaging data and the final diagnosis. The neural network processes the imaging data through multiple layers, extracting features and generating malignancy likelihood scores, thereby bridging the gap between complex 3D data and actionable medical insights.
2Measurement precision
If a neural network is trained on multiple types of annotated images (patches, bounding boxes, image-level labels), then detection accuracy is improved, but training complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-training the neural network on annotated patches and bounding boxes before final training on image-level labeled data. This staged approach allows the model to learn hierarchical features from simple to complex representations, reducing overall training complexity while improving accuracy.
Solution Approach 2:
The patent changes training parameters dynamically during the training process, transitioning from patch-level to image-level annotations. The system adjusts the level of annotation detail and training objectives based on the current training stage, thereby managing complexity while improving detection accuracy.
3Loss of time
If 2D imaging is used for mammography, then analysis time is reduced, but detail and diagnostic capability are insufficient
Solution Approach 1:
The patent transitions from 2D imaging to 3D tomosynthesis, adding the temporal/dimensional dimension of multiple angled views. This allows the system to provide enhanced diagnostic detail by reconstructing 3D breast anatomy from multiple 2D projections, effectively adding information without requiring a complete paradigm shift.
Data Source
AI summary
The present disclosure provides a computerized method for a likelihood of malignancy in breast tissue of a patient. The method includes receiving, with a computer processor, an image of the breast tissue, providing the image of the breast tissue to a model including a trained neural network; the trained neural network being previously trained by training a first neural network, initializing a second neural network based on the first neural network, training the second neural network, and outputting the second neural network as the trained neural network, receiving an indicator from the model, and outputting a report including the indicator to at least one of a memory or a display.