Digital Image Quality Assessment for Anatomical Content Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated image quality assessment systems for medical imaging struggle to accurately evaluate image content quality, particularly in cases of misalignment or incorrect anatomical content, leading to misjudgment by radiographers and increased costs due to retakes.
Innovation Solution
A deep-learning model that combines multi-resolution convolutional layers and bi-directional feature pyramid networks to assess image content quality, providing a score based on body part and view position information, trained on multiple acquisition types to ensure accurate detection of required anatomical content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If automated image quality assessment systems are implemented, then image quality evaluation accuracy is improved, but system complexity and training requirements increase
Solution Approach 1:
The patent implements a universal deep learning model that can assess multiple acquisition types (e.g., thorax PA, thorax lateral, shoulder AP, abdomen AP, knee AP, knee lateral, hand PA, hand lateral, foot AP, foot lateral) simultaneously. The model is trained on diverse datasets covering various body parts and projection types, enabling it to generalize across different examination protocols without requiring separate specialized systems for each acquisition type.
Solution Approach 2:
The assessment system divides image quality evaluation into distinct components: technical image quality parameters (noise, contrast, sharpness) and image content quality (presence and visibility of anatomical features). This segmentation allows the model to evaluate different aspects independently and combine them into an overall quality score, making the complex assessment process more manageable and interpretable.
2Measurement precision
If radiographers perform manual image quality assessment, then image content quality can be evaluated, but time consumption and human error increase
Solution Approach 1:
The patent replaces the manual visual assessment process (mechanical system involving human radiographers) with an automated deep learning-based computer vision system. The model processes images algorithmically to evaluate technical parameters and anatomical content quality, eliminating human subjectivity and fatigue while maintaining or improving assessment accuracy. This substitution enables rapid automated evaluation without the time constraints and human errors associated with manual review.
3Reliability
If image retakes are performed to ensure quality, then image quality is improved, but patient dose and cost increase
Solution Approach 1:
The patent implements preliminary automated quality assessment immediately after image acquisition, evaluating both technical parameters and anatomical content quality before the image is used for diagnosis. This preliminary check identifies suboptimal images that require retake, allowing quality assurance to be performed proactively. By detecting issues early while the patient is still present and positioned, the system prevents the need for additional radiation exposure from repeat examinations.
Solution Approach 2:
The system provides immediate feedback on image quality through automated assessment results, indicating whether the acquired image meets quality criteria for diagnostic use. This feedback mechanism enables real-time quality control, allowing radiographers to make informed decisions about whether an image is acceptable or requires retake, thereby optimizing patient dose by avoiding unnecessary repeat exposures.
Data Source
AI summary
The invention relates to a method to give an indication about the image quality of a digital image in comparison to what the expected image quality in terms of image content and technical image quality parameters would be for a similar exposure type. The method evaluates whether parameters of the acquired image such as noise and dynamic range match the expectations for the intended exposure type, and whether certain regions of interest are present and properly presented in the image.