Upstream Data Fusion for Abnormal Tissue Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current breast cancer detection methods face limitations in accuracy due to high false positives and lengthy read times, largely due to the inability to effectively combine and process substantial amounts of medical imaging data from various modalities, leading to missed diagnoses and unnecessary procedures.
Innovation Solution
The implementation of an upstream data fusion (UDF) approach that combines data from multiple medical imaging modalities using machine learning to generate feature likelihood models, reduce false positives, and automate the detection of abnormal tissue with confidence scores and anatomic locations, thereby reducing radiologist read times and false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple imaging modalities are used to improve detection accuracy, then the amount of data increases, but radiologist read time increases and false positives remain high
Solution Approach 1:
The patent introduces an automated data fusion system as an intermediary between multiple imaging modalities and radiologist interpretation. This system processes and integrates data from tomosynthesis, MRI, and ultrasound modalities automatically, generating fused images and detection candidates without requiring radiologists to manually review all raw data, thereby maintaining high detection accuracy while significantly reducing read time
Solution Approach 2:
The patent segments the complex task of multi-modal image analysis into distinct processing stages: individual modality processing, data fusion, detection candidate generation, and prioritization. Each stage handles specific computational tasks independently, allowing parallel processing and optimizing the overall workflow efficiency, which reduces radiologist read time while preserving detection accuracy
2Measurement precision
If multiple imaging modalities are used to improve detection accuracy, then more data is available, but false positive rate increases leading to unnecessary procedures
Solution Approach 1:
The patent implements feedback mechanisms where detection candidates are generated with associated confidence scores and prioritized based on fusion results. The system provides feedback loops that allow iterative refinement of detection algorithms and threshold optimization, enabling the system to learn from false positives and improve specificity while maintaining high sensitivity for true cancer detection
Solution Approach 2:
The patent dynamically adjusts detection parameters and thresholds based on the fused data from multiple modalities. By changing detection parameters adaptively according to the quality and consistency of data across different modalities, the system optimizes the balance between sensitivity and specificity, reducing false positives while maintaining high detection accuracy
3Reliability
If comprehensive imaging data is reviewed to reduce missed diagnoses, then detection thoroughness improves, but read time increases significantly
Solution Approach 1:
The patent performs preliminary automated processing and fusion of imaging data before radiologist review. Detection algorithms pre-process the multi-modal data to generate prioritized detection candidates, allowing radiologists to focus their expertise on evaluating pre-identified suspicious areas rather than reviewing all images from scratch, thereby maintaining high reliability while improving screening throughput
4Productivity
If automated data fusion is implemented to reduce read time, then processing speed increases, but system complexity increases
Solution Approach 1:
The patent implements a universal data fusion architecture that can process multiple imaging modalities (tomosynthesis, MRI, ultrasound) through a common processing framework. This multi-functional system uses standardized interfaces and algorithms that can handle different data types, reducing the need for separate processing pipelines for each modality and managing system complexity while maintaining high processing speed
Data Source
AI summary
A method for characterizing and locating abnormal tissue in an area of interest in a body is provided. The method may include receiving first and second modality examination data including a plurality of first and second modality examination data segments of the area of interest and determining, by evaluating each first and second modality examination data segment, a set of modality detection candidates based on the first and second modality examination data and a first and second modality feature likelihood model. The method may also include generating a modal or multi-modal correspondence anatomic model of the area of interest by registering the plurality of first and second modality examination data segments, determining one or more modal abnormal tissue detections by fusing the set of modality detection candidates using the modal or multimodal correspondence anatomic model, and providing a modal diagnosis confidence score and a modal anatomic location based on the one or more modal abnormal tissue detections.