Automated Aneurysm Boundary Detection and Volume Estimation Using Deep Learning.

Area of Science:

• Medical Imaging Analysis
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Current aneurysm assessment relies on manual imaging review, which is time-consuming and variable.
• Standard care measures only the widest diameter, limiting morphological insights.
• Accurate volume measurement is crucial for understanding aneurysm progression and severity.

Purpose of the Study:

• To develop an automated method for precise aneurysm boundary delineation and volume computation.
• To leverage pre-trained neural networks and expert systems for aneurysm analysis.
• To validate the approach on an unannotated dataset of 60 patients.

Main Methods:

• Utilized a pre-trained UNet for automatic aorta localization.
• Employed SAM2 to track the aorta through complex vascular structures, including aneurysms.
• Applied a Long Short-Term Memory (LSTM) network or expert system to identify aneurysm start/end points.

Main Results:

• Achieved high accuracy in predicting aneurysm start (R²=71%) and end points (R²=76%).
• Demonstrated excellent accuracy in aneurysm volume prediction with an R² score of 92%.
• Successfully computed volumes without manual aorta segmentation.

Conclusions:

• The automated technique shows significant potential for large-scale aneurysm analysis.
• This method can enhance clinical decision-making by reducing reliance on annotated datasets.
• Automated volume measurement offers a more comprehensive understanding of aneurysm morphology and risk.