Clinical Decision Support via Context-Specific Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current clinical decision support systems rely on image features extracted from imaging data optimized for human reading, which are suboptimal for computer-based analysis, leading to approximate and coarse similar case retrieval, limiting decision support effectiveness.
Innovation Solution
A method and system that generate preliminary imaging data optimized for feature extraction, extract clinical context-specific features, annotate them for efficient storage and retrieval, and select similar case data sets from a reference database based on annotated features for enhanced clinical decision support.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If image features are extracted from imaging data optimized for human reading (thick slices, smooth reconstruction kernels), then fast network transmission and PACS storage are achieved, but feature extraction quality for computer-based analysis becomes suboptimal
Solution Approach 1:
The patent segments the imaging data processing into two distinct pathways: one for human reading (thick slices, smooth kernels) and another for computer-based analysis (thin slices, sharp kernels). This segmentation allows each pathway to be optimized independently, resolving the contradiction between transmission speed and feature extraction precision.
Solution Approach 2:
The patent applies local quality by using different reconstruction kernels and slice thicknesses for different purposes. For computer-based analysis, thin slices with sharp reconstruction kernels are used to maximize feature extraction quality, while human reading uses thick slices with smooth kernels. This local optimization resolves the contradiction by tailoring data characteristics to specific use cases.
2Quantity of substance
If image data resolution is reduced to decrease data size for faster network transmission and PACS storage, then transmission efficiency is improved, but feature extraction quality for characterizing lesions and lung parenchyma deteriorates
Solution Approach 1:
The patent segments image data into two representations: compressed data for PACS storage and transmission, and full-resolution data for feature extraction. This segmentation allows the system to maintain high-quality features for analysis while keeping PACS data sizes manageable, resolving the contradiction between data size and feature extraction precision.
Solution Approach 2:
The patent performs feature extraction on high-resolution data before it is sent to PACS or reading workstations. By extracting features in advance from the original high-resolution images, the system preserves feature extraction precision while allowing PACS to store only the compressed versions, thus resolving the contradiction.
3Quantity of substance
If similar case retrieval is based on a very large number of patients shared across institutions using cloud computing, then the comprehensiveness of the patient population data pool is improved, but data transmission and processing complexity increases
Solution Approach 1:
The patent extracts only the essential feature data from medical images, converting complex imaging data into compact feature representations (such as hash codes). This extraction reduces the complexity of data transmission and processing while maintaining the ability to perform accurate similar case retrieval across large patient populations.
Solution Approach 2:
The patent transforms complex medical image data into simplified feature parameters (e.g., hash codes, extracted features) that preserve the essential information needed for similarity comparison. This parameter transformation reduces data complexity and transmission requirements while enabling efficient retrieval across large distributed databases.
Data Source
Figure 1
Figure 2
AI summary
The invention provides a method for supporting clinical decisions for a diagnosis or therapy of a patient using a medical imaging system comprising the steps of, receiving a procedure order; based on the received procedure order, automatically identifying a clinical context of the ordered procedure; generating preliminary imaging data of at least a part of the patients anatomy; generating feature extraction data based on the identified clinical context and on the preliminary imaging data data; extracting at least one clinical context specific feature using the generated feature extraction data; annotating the at least one extracted clinical context specific feature to obtain at least one annotated extracted feature; for the identified clinical context, selecting a similar case data set from a reference database of case data sets based on the at least one annotated extracted feature.