A Deep Learning System For The Automated Detection And Segmentation Of Abdominal Aortic Aneurysm From Computed Tomography Angiography
Keywords:
Abdominal Aortic Aneurysm, Deep Learning, Image Segmentation, Computed Tomography Angiography, Convolutional Neural Networks, nnU-Net.Abstract
Abdominal Aortic Aneurysm (AAA) is a life-threatening condition whose management relies on the precise measurement from Computed Tomography Angiography (CTA), a process hindered by the time-consuming and variable nature of manual segmentation. To address this, we developed and validated a deep learning system for the fully automated detection and segmentation of AAA. Using a retrospective dataset of [Number] CTA studies, we trained a convolutional neural network based on the nnU-Net framework to perform slice-level detection and voxel-level segmentation of the aortic lumen and thrombus. On an independent test set, our system achieved a mean Dice Similarity Coefficient of [e.g., 0.95 ± 0.03] for the lumen and [e.g., 0.88 ± 0.06] for the thrombus, and demonstrated exceptional detection accuracy of [e.g., 99.1%] with excellent agreement between automated and manual diameter measurements (intraclass correlation coefficient [e.g., 0.98]). This robust system promises to streamline clinical workflow, reduce radiologist workload, and provide rapid, reproducible AAA assessments to enhance clinical decision-making.
