Linden, T. van der (Thijs) (2019) Determining aortic area and distensibility using deep learning on CMR images. thesis, Medicine.
Full text available on request.Abstract
Introduction: Decreased ascending aortic distensibility (AAD) is strong predictor of future cardiovascular events. AAD can be determined by manually measuring the ascending aortic area (AAA) on cardiovascular magnetic resonance images. However, this is a time consuming process and prone to subjective bias, which is undesired, especially in large cohorts. The aim of this study was to investigate the use of deep learning in automatically assessing AAA and AAD. Methods: First, a deep learning model was developed on the Cosmonio platform (Groningen, the Netherlands) using an annotated training set of 3000 images. Next, the model was used to asses AAD and AAA in 192 participants from the UK biobank (54% male, mean age 62±7.6 years) during a whole cardiac cycle (100 frames). The measurements derived from the deep learning model were compared to a group of manual measurements. Agreement between automated and manual measurements was evaluated using the dice similarity coefficient, the Pearson’s correlation coefficient, and Bland-Altman plots. Results: AAD was 2.4 and 2.6 × 10-3 mm Hg-1 in females and males respectively (P=0.484). Maximum AAA was 10.7 and 12.6 for females and males respectively (P<0.001). The deep learning model achieved a dice coefficient of 0.91. Bias between automated and manual AAA was 3.15 cm2 (95% limits of agreement 1.10-5.17 cm2). The correlation coefficient was 0.86. Conclusion: Although a consistent over-prediction by the deep learning model was observed, high correlation with manually derived aortic measurements was achieved. Automated analysis of the aorta to derive area and distensibility measurements is promising for time-efficient and reproducible large scale measurements which can be used for individual risk prediction of cardiovascular disease.
| Item Type: | Thesis (Thesis) |
|---|---|
| Supervisor name: | Hendriks, drs. Tom and Harst, Prof dr. Pim van der and Department and Cardiology, University Medical Center Groningen |
| Faculty: | Medical Sciences |
| Date Deposited: | 25 Jun 2020 11:02 |
| Last Modified: | 25 Jun 2020 11:02 |
| URI: | https://umcg.studenttheses.ub.rug.nl/id/eprint/2247 |
Actions (login required)
 |
View Item |