Fully Automated Lipid Pool Detection Using Near Infrared Spectroscopy.
Elżbieta PociaskJoanna Krystyna Jaworek-KorjakowskaKrzysztof Piotr MalinowskiTomasz RolederWojciech WojakowskiPublished in: Computational and mathematical methods in medicine (2016)
Background. Detecting and identifying vulnerable plaque, which is prone to rupture, is still a challenge for cardiologist. Such lipid core-containing plaque is still not identifiable by everyday angiography, thus triggering the need to develop a new tool where NIRS-IVUS can visualize plaque characterization in terms of its chemical and morphologic characteristic. The new tool can lead to the development of new methods of interpreting the newly obtained data. In this study, the algorithm to fully automated lipid pool detection on NIRS images is proposed. Method. Designed algorithm is divided into four stages: preprocessing (image enhancement), segmentation of artifacts, detection of lipid areas, and calculation of Lipid Core Burden Index. Results. A total of 31 NIRS chemograms were analyzed by two methods. The metrics, total LCBI, maximal LCBI in 4 mm blocks, and maximal LCBI in 2 mm blocks, were calculated to compare presented algorithm with commercial available system. Both intraclass correlation (ICC) and Bland-Altman plots showed good agreement and correlation between used methods. Conclusions. Proposed algorithm is fully automated lipid pool detection on near infrared spectroscopy images. It is a tool developed for offline data analysis, which could be easily augmented for newer functions and projects.
Keyphrases
- deep learning
- machine learning
- convolutional neural network
- artificial intelligence
- data analysis
- fatty acid
- loop mediated isothermal amplification
- coronary artery disease
- real time pcr
- optical coherence tomography
- label free
- high throughput
- heart rate
- big data
- risk factors
- neural network
- resistance training
- quantum dots
- sensitive detection
- image quality