Vascular and Intravascular Imaging Trends, Analysis, and Challenges, Volume 2: Plaque characterization by Jasjit S Suri Petia Radeva

Author:Jasjit S Suri Petia Radeva [Petia Radeva, Jasjit S Suri]
Language: eng
Format: mobi
ISBN: 9780750320023
Publisher: IOP Publishing
Published: 2019-08-04T16:00:00+00:00

5.6 Conclusion

Our study showed the application of deep learning for automated lumen detection and LD measurement, using a class of AtheroEdge™ systems from AtheroPoint™. The DL system used a multiresolution strategy in a DL framework. The system consisted of 13 layers for feature extraction in stage one and three layers in stage two of the FCN network. Three expert readings were taken for evaluation. Using the metric of polyline distance, the PoM estimation between the manual and DL system over 407 US scans was 99.61%, 97.75% and 99.89%, respectively. The Jaccard index and Dice similarity of the DL lumen segmented region against three GT regions were 94%, 94% and 93%, and 97%, 97% and 97%, respectively. Using the threshold of 0.85 mm, the AUC obtained from ROC analysis showed an accuracy of 0.95, 0.91 and 0.93, corresponding to three manual readings. Benchmarking the DL system against the previously published literature showed an improvement in LD error, and LI-far and LI-near wall boundary errors of 24%, 33% and 4%, respectively. The new and interesting conclusions can be described briefly as follows. First, the intelligent-based deep-learning strategy provides a more powerful tool of lumen segmentation compared to conventional image processing methods. The capture of global shape using the knowledge derived from the gold standard is a powerful guide during the cross-validation schemes in learning infrastructure. Second, the deep-learning paradigm, a specific class under the intelligent-based infrastructure, when adapted in the multiresolution paradigm can lead to effective feature extraction during deep learning due to noise suppression in the down-sampling mode. Third, since the variability study shows better statistical correlations and successful passing of statistical tests with strong p values (< 0.0001), this clearly demonstrates the power of the deep-learning strategies for lumen risk stratification and characterization. Fourth, our benchmarking evaluation covering the last 15 years of published research shows that deep learning yields the best performance in terms of reliability, accuracy and stability—as demonstrated in this chapter. The deep-learning system is reliable, the best in the class of AtheroEdge™ systems and can be adapted for clinical settings.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.