Academic Journal
Abstract 4146625: Automated Detection and Measurement of Aortic Aneurysms in a Clinical Population
| Τίτλος: | Abstract 4146625: Automated Detection and Measurement of Aortic Aneurysms in a Clinical Population |
|---|---|
| Συγγραφείς: | Ian Dinsmore, Jonathan Luo, William Triffo, Evan Ryer, James Elmore, David Carey, Tooraj Mirshahi |
| Πηγή: | Circulation. 150 |
| Στοιχεία εκδότη: | Ovid Technologies (Wolters Kluwer Health), 2024. |
| Έτος έκδοσης: | 2024 |
| Θεματικοί όροι: | 03 medical and health sciences, 0302 clinical medicine |
| Περιγραφή: | Introduction: Aortic aneurysms (AA*) are enlargements of the aorta that can be life-threatening if left untreated. Early detection is essential and typically relies on imaging. However, routine screening is only performed for at-risk groups, such as older male smokers. Computed tomography (CT) scans performed for other reasons could be used to enhance early AA* detection. This could lead to significantly increased screening rates and reduce AA* related deaths. Hypothesis: A subgroup of the clinical population who would not undergo routine screening for AA* will have a CT showing aortic ectasia or an aortic aneurysm. Goals: To automatically measure aortic diameters from routine CT imaging on a large scale and provide early AA* detection. Methods: We trained a deep-learning model on a public dataset of 982 CT scans split into 933 training/validation samples and 49 independent testing samples. We segmented the aorta in both non-contrast and contrast CT images and applied real-time image augmentation during training to enhance the data. The aorta segmentation model achieved a Dice Similarity Coefficient of 0.96 during training, and 0.95 during testing. Using an automated centerline analysis, we measured the maximum cross-sectional diameter of the thoracic (ascending and descending), and abdominal aorta. Results: We extracted 34,805 unique abdominal and 26,629 unique chest CTs and measured aortic diameter in each scan. Men had significantly larger abdominal and thoracic aortic diameters than women (p-value < 1e-26, unpaired t-test): ascending: 37.9±4.3 vs 34.2±4.0mm; descending: 31.5±4.2 vs 27.8±3.7mm; abdominal: 24.4±7.0 vs 20.0±4.1mm (men vs women, mean±SD). We manually reviewed 51 ascending, 51 descending, and 91 abdominal aorta diameters. We compared the diameters to the automatic measurements, finding mean absolute errors of 1.6mm, 1.0mm, and 2.8mm, respectively. Evaluating this cohort against ICD10 codes showed that out of 1,362 patients with an abdominal diameter > 3cm, 277 (20%) did not have a AAA diagnosis. Out of 1,126 patients with a descending diameter > 4cm or an ascending diameter > 4.5cm, 386 (34%) did not have a TAA diagnosis. Using existing CT images, we identified a large number of subjects who may be at risk for AA* but have not received routine screening for the disease. Conclusions: By using a deep-learning segmentation model on CT imaging, we can automatically and accurately measure aortic diameter and detect AA*. |
| Τύπος εγγράφου: | Article |
| Γλώσσα: | English |
| ISSN: | 1524-4539 0009-7322 |
| DOI: | 10.1161/circ.150.suppl_1.4146625 |
| Αριθμός Καταχώρησης: | edsair.doi...........d4684e3fb8f73a2f9df08a52f68753de |
| Βάση Δεδομένων: | OpenAIRE |
| ISSN: | 15244539 00097322 |
|---|---|
| DOI: | 10.1161/circ.150.suppl_1.4146625 |