Medical Image Artifact Localization Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Medical images often contain artifacts that can confuse radiologists, leading to inaccurate diagnoses and improper treatments, as they cannot be reliably distinguished from actual abnormalities.
Innovation Solution
A system using machine learning models to identify, classify, and localize artifacts in medical images, employing techniques such as convolutional neural networks and post-processing to highlight relevant artifacts on a graphical user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are used to identify and localize artifacts in medical images, then diagnostic precision is improved, but device complexity increases
Solution Approach 1:
The system segments the artifact detection task into multiple specialized machine learning models, each trained to detect specific types of artifacts (e.g., motion artifacts, metal artifacts, streak artifacts). This segmentation allows the complex problem of artifact detection to be divided into manageable components, improving diagnostic precision while organizing system complexity in a structured manner
Solution Approach 2:
The patent introduces an intermediary processing layer that includes post-processing techniques such as localization algorithms and artifact scoring mechanisms. These intermediaries bridge the gap between raw machine learning predictions and clinically actionable insights, enhancing diagnostic precision by providing radiologists with localized artifact information and confidence scores without requiring them to interpret raw model outputs
2Measurement precision
If multiple machine learning models are employed to detect different artifact types, then artifact detection accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary classification to identify which types of artifacts are present in an image before applying specialized detection models. This preliminary action allows the system to selectively apply only the necessary machine learning models for each image, improving artifact detection accuracy while reducing unnecessary processing time for images with fewer artifact types
Solution Approach 2:
The patent implements a tiered detection approach where a lightweight initial screening model quickly identifies images containing artifacts. For images flagged by the screening model, more computationally intensive specialized models are applied. This partial action strategy ensures high detection accuracy for artifact-containing images while minimizing processing time for artifact-free images
3Measurement precision
If artifact localization techniques are applied to highlight artifacts, then differentiation between artifacts and abnormalities is improved, but ease of operation decreases
Solution Approach 1:
The system applies local quality enhancement by localizing artifacts to specific regions of the medical image and providing location-specific information to radiologists. Rather than treating the entire image uniformly, the system highlights only the regions containing artifacts with localized bounding boxes, masks, or annotations. This approach improves differentiation accuracy by drawing attention to specific problem areas while maintaining ease of operation by not requiring radiologists to analyze the entire image for artifacts
4Reliability
If comprehensive artifact classification is implemented, then diagnostic reliability is improved, but loss of information increases
Solution Approach 1:
The patent implements feedback mechanisms that provide radiologists with confidence scores, classification probabilities, and localization confidence metrics alongside artifact detections. This feedback loop allows radiologists to assess the reliability of each artifact detection and selectively review cases where the system's confidence is lower, improving diagnostic reliability while preserving information by allowing human judgment to supplement automated classifications
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
A method (500), device (104) and system (100) for generating medical image with localized artifacts are provided. In one aspect, the method (500) includes receiving, by a processing unit (202), a medical image generated by an imaging modality (102A-N), identifying presence of one or more artifacts in the medical image using one or more trained machine learning models, determining at least one artifact of interest from the one or more artifacts based on a predefined criteria, localizing the at least one artifact of interest in the medical image using a post processing technique, and outputting the medical image comprising localized artifact of interest on a graphical user interface (900).