Author + information
- Taylor Hoyt1,
- Vikram Baruah2,
- Aydin Zahedivash3,
- Andrew Cabe4,
- Nathanael Phillips4 and
- Marc Feldman5
- 1Clayton Research Foundation, San Antonio, Texas, United States
- 2University of Texas at Austin, Austin, Texas, United States
- 3Dell Medical School, University of Texas, Austin, Texas, United States
- 4UTHSCSA, San Antonio, Texas, United States
- 5University of Texas Health Science Center at San Antonio, San Antonio, Texas, United States
Intravascular optical coherence tomography use in the cardiac catheterization laboratory setting has been limited, in part due to the complexity of image interpretation. TCFA can be falsely identified due to tissue morphologies that mimic them.
Fifty-four human coronary arteries were imaged ex vivo with a St. Jude ILUMIEN IVOCT system and then stained with H&E. After coregistration, regions of interest (ROI) were selected from the IVOCT and categorized as fibrous, calcium, lipid and necrosis. The necrosis category was constructed exclusively from ROIs of necrotic cores of TCFAs to better detect this lesion. Texture features were extracted from the ROIs and then input into tissue-specific neural networks. Additionally, values deduced from the A-scan profile of the image were used to weight pixels.
When the networks are trained on a set of pixels, and then tested on different pixels in the same pullbacks, fibrous, calcium, lipid, and necrosis pixels were classified with 96, 93, 94, 88% accuracy, respectively. Accuracies for neural networks on independent arteries without retraining are 87, 82, 86, and 81%.
A histology-validated IVOCT-based automated plaque classification algorithm has been developed to colorize plaque composition in human coronary arteries. With minimal expert assistance, accuracies reported herein exceed previously reported values for pixel wise automated IVOCT plaque classification. Histological validation makes the presented method more reliable than expert observer-validated approaches.
IMAGING: Imaging: Intravascular