Deep Learning Motion Detection for PET/CT Misalignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Nuclear imaging systems face misalignment issues due to subject movement during image capture, leading to errors in reconstruction and clinical interpretation of PET and CT scans.
Innovation Solution
A deep learning-based approach using two trained neural networks to detect and quantify inter-modal movement between PET and CT images, generating displacement data for improved clinical interpretation, including heat maps and alerts for significant misalignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional image reconstruction methods are used, then the reconstruction process is simple and fast, but spatial misalignment between PET and CT images occurs due to subject movement
Solution Approach 1:
The system performs motion detection and quantification before final image reconstruction and clinical interpretation. By detecting displacement between PET and CT images using trained neural networks, the system identifies misalignment issues upfront, allowing for appropriate corrective actions or rescanning decisions before reconstruction errors are locked in.
Solution Approach 2:
The patent introduces an intermediary motion detection system that analyzes displacement data between PET and CT images. This intermediary layer uses trained neural networks to generate motion indicators and heat maps, serving as a bridge between raw imaging data and final reconstructed images, thereby ensuring spatial consistency without directly modifying the reconstruction process.
2Measurement precision
If deep learning-based motion detection is implemented, then motion detection accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The neural networks are trained in advance on large datasets of paired PET and CT images with known displacements. This preliminary training phase allows the system to achieve high motion detection accuracy during actual clinical use without performing complex computations in real-time, thereby reducing processing time while maintaining precision.
Solution Approach 2:
The system uses trained neural network models that have learned motion patterns from training data. Instead of performing complex motion analysis from scratch on each new image pair, the system applies pre-trained models that have captured the essential motion characteristics, significantly reducing processing time while maintaining detection accuracy.
Data Source
AI summary
Systems and methods for detecting subject motion within medical images based on trained deep learning processes are disclosed. In some examples, an image processing system receives positron emission tomography (PET) measurement data and co-modality measurement data from an image scanner. The image processing system generates PET images and co-modality images based on the PET measurement data and co-modality measurement data, respectively. Further, the image processing system inputs the PET images and the co-modality images to a first trained neural network, and generates first features of the PET measurement data and second features of the co-modality measurement data. The image processing system inputs the first features and the second features to a second trained neural network and, generates displacement data characterizing a displacement between the first features and the second features. Based on the displacement data, the image processing system generates display data for display.