Protect the Brain When Treating the Heart: A Convolutional Neural Network for Detecting Emboli
A research paper on arXiv proposes a 2.5D U-Net architecture to detect gaseous microemboli (GME) in cardiac ultrasound imaging. GME are a common complication of structural heart interventions. The model segments GME in space-time data, achieving robust detection and high accuracy while maintaining real-time speed, enabling integration into patient-monitoring protocols.
Key facts
- Gaseous microemboli (GME) are a complication of cardiac structural interventions.
- Transthoracic cardiac ultrasound is used to visualize GME.
- Detection is difficult due to operator-dependent views, high velocity, and background objects.
- A 2.5D U-Net architecture is proposed for GME segmentation.
- The approach yields robust detection and high segmentation accuracy.
- Real-time execution speed is maintained.
- The pipeline is integrated into patient-monitoring surgical protocols.
- Quantification of GME area over time is provided.
Entities
Institutions
- arXiv