Related Experiment Video
Updated: Jul 11, 2025
Improved Registration of 3D CT Angiography with X-ray Fluoroscopy for Image Fusion During Transcatheter Aortic Valve Implantation
Published on: June 3, 2018
DLFFNet: A new dynamical local feature fusion network for automatic aortic valve calcification recognition using
Lingzhi Tang1, Xueqi Wang2, Jinzhu Yang1
1Key Laboratory of Intelligent Computing in Medical Image, Ministry of Education, Northeastern University, China; Computer Science and Engineering, Northeastern University, Shenyang, China; National Frontiers Science Center for Industrial Intelligence and Systems Optimization, China.
A new deep learning model automates aortic valve calcification (AVC) detection in echocardiograms, improving accuracy and efficiency over manual methods. This AI tool aids cardiovascular event prediction by reliably identifying AVC from ultrasound images.
Area of Science:
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
- Medical Image Analysis
Background:
- Aortic valve calcification (AVC) is a key predictor of cardiovascular events, often linked to coronary artery stenosis.
- Current AVC identification via echocardiography relies on subjective visual assessment, leading to inter-observer variability and extensive training requirements.
- Accurate AVC detection in echocardiographic images is challenging due to image artifacts and anatomical variations.
Purpose of the Study:
- To develop an automated system for accurate aortic valve calcification (AVC) identification using echocardiographic images.
- To overcome the limitations of manual AVC assessment, including subjectivity and time-intensive learning curves.
- To enhance the reliability and efficiency of AVC detection for improved cardiovascular risk stratification.
Main Methods:
- A dynamical local feature fusion network was developed for automated AVC recognition from echocardiograms.
- High-echo areas were segmented using U-Net, with fine-tuning via a mask tuning module for dynamic local feature selection.
- A pyramid-based, two-branch feature fusion module was designed to integrate multi-level, multi-scale, global, and local semantic information.
- A unified preprocessing algorithm was implemented to address variations in aortic valve position and size across different datasets.
Main Results:
- The proposed model achieved high performance on a dataset of 231 patients' echocardiographic images.
- Key performance metrics included accuracy (82.40%), precision (82.50%), sensitivity (82.50%), specificity (91.23%), and F1 score (82.47%).
- The model demonstrated strong discriminative power with micro-AUC of 92.39% and macro-AUC of 92.25%.
Conclusions:
- The study demonstrates the feasibility of automated AVC examination using echocardiography.
- Visualization confirmed that the AI model's regions of interest align with expert interpretations.
- The developed AI approach offers a promising tool for objective and efficient AVC assessment in clinical practice.
