Abstract
Cardiovascular disease (CVD) is the leading cause of death in the United States, yet it is largely preventable. But a critical part of prevention is identification of at-risk persons before adverse events. For predicting individual CVD risk, carotid intima-media thickness (CIMT), a noninvasive ultrasonography method, has proven to be valuable. However, each CIMT examination includes several ultrasonography videos, and interpreting each CIMT video involves 3 operations: (i) selecting 3 end-diastolic ultrasonographic frames (EUFs) in the video, (ii) localizing a region of interest (ROI) in each selected frame, and (iii) tracing the lumen-intima interface and the media-adventitia interface in each ROI to measure CIMT. These operations are tedious, laborious, and time-consuming, hindering the widespread utilization of CIMT in clinical practice. We present a new system based on convolutional neural networks to automate the entire process of CIMT video interpretation. The suggested system achieves a mean absolute error of 23.4 μm with a standard deviation of 17.3 μm for intima-media thickness measurements. The ANOVA test also yields p-values around 0.50, suggesting that there is a lack of evidence to show a difference among the CIMT measurements made by our system and those made by 3 experts. Our results suggest that the proposed system is robust against variability in CIMT measurements made by different experts.
Original language | English (US) |
---|---|
Title of host publication | Deep Learning for Medical Image Analysis |
Publisher | Elsevier Inc. |
Pages | 105-131 |
Number of pages | 27 |
ISBN (Electronic) | 9780128104095 |
ISBN (Print) | 9780128104088 |
DOIs | |
State | Published - Jan 30 2017 |
Keywords
- Cardiovascular disease risk stratification
- Carotid intima-media thickness
- Convolutional neural networks
- Deep learning
ASJC Scopus subject areas
- General Engineering