Deep Learning LVO Detection in CTA Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting Large Vessel Occlusion (LVO) in Computational Tomography Angiogram (CTA) images are inefficient due to noise artifacts from delayed acquisition, leading to inaccurate and complicated processing that often requires human intervention.
Innovation Solution
A system and method utilizing deep learning models, including a cranium classifier, vascular territory segmentation, ICV segmentation, MCA-LVO classifier, and ICA-LVO classifier, to automatically detect LVO by removing noise artifacts and constructing noise-free Maximum Intensity Projections (MIP) for accurate segmentation and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If CTA images are acquired quickly, then the detection time for LVO is reduced, but noise artifacts increase due to contrast agents reaching veins
Solution Approach 1:
The system extracts and removes noise artifacts from CTA images using deep learning-based denoising algorithms. The neural network identifies and eliminates venous contrast artifacts while preserving arterial structures, enabling fast acquisition without compromising image quality for LVO detection
Solution Approach 2:
The system changes the temporal parameters of image acquisition to optimize the balance between speed and quality. By adjusting the timing window for image capture and using rapid processing algorithms, the system achieves sub-5-minute LVO detection while managing contrast agent distribution to minimize venous noise
2Measurement precision
If deep learning models are used for automatic LVO detection, then the accuracy of detection is improved, but the computational complexity increases
Solution Approach 1:
The deep learning system is segmented into specialized modules: a cranium classifier for skull stripping, a vascular territory segmentation module for anatomical region identification, and an LVO detection module for occlusion identification. This modular architecture improves accuracy while managing computational complexity through targeted processing in each segment
Solution Approach 2:
The system performs preliminary actions by pre-processing CTA images through skull stripping and vascular segmentation before LVO detection. The cranium classifier removes skull artifacts in advance, and vascular territories are pre-segmented, reducing the computational burden on the final LVO detection step while maintaining high accuracy
3Measurement precision
If manual processing of CTA images is performed, then the accuracy of LVO detection is improved, but the processing time increases and human intervention is required
Solution Approach 1:
The system implements self-service through automated deep learning-based LVO detection that processes CTA images without human intervention. The neural networks automatically perform skull stripping, vascular segmentation, and occlusion detection, achieving both high accuracy and rapid processing speeds, eliminating the trade-off between manual accuracy and automated speed
Data Source
AI summary
The present subject matter discloses a system and method for detecting Large Vessel Occlusion (LVO) on a Computational Tomography Angiogram (CTA) automatically. the system comprises a vascular-territory-segmentation module, an ICV segmentation module, MCA-LVO classifier and ICA-LVO classifier. The vascular territory segmentation module is configured to receive a set of CTA images and to mark a territory of vascular segments in the ICV region for each slice of the ROI. The ICV segmentation module is configured to process each slice of the ROI. The processed slices of the ROI are combined to develop a CTA image after application of MIP and the developed CTA image is segmented into a Middle Cerebral Artery (MCA) region and an Internal Cerebral Artery (ICA) region. The MCA-LVO and ICA-LVO classifiers determine presence of the LVO on the received MCA and ICA region using Deep Learning techniques and accordingly the presence of the LVO is reported.