Automatic detection of intracranial aneurysms in 3D-DSA based on a Bayesian optimized filter

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Cerebrovascular Diseases

Background:

• Intracranial aneurysms pose a risk of subarachnoid hemorrhage.
• Accurate detection is crucial for timely intervention.
• 3D digital subtraction angiography (DSA) offers superior vascular detail compared to 2D imaging.

Purpose of the Study:

• To develop an adaptive multiscale filter for detecting intracranial aneurysms.
• To improve diagnostic accuracy using 3D-DSA.
• To reduce the misdiagnosis rate in cerebrovascular disease diagnosis.

Main Methods:

• Utilized a Hessian matrix eigenvalue-based filter with Bayesian optimization.
• Employed region growing and adaptive thresholding for aneurysm extraction.
• Implemented an iterative detection strategy for identifying multiple aneurysms.

Main Results:

• Achieved a detection precision of 94.6% and sensitivity of 96.4% on a dataset of 145 patients.
• Successfully detected 93.9% of aneurysms smaller than 5 mm.
• Demonstrated superior performance over 2D-DSA and comparable or better results than other modalities.

Conclusions:

• The proposed adaptive multiscale filter is stable and reliable for intracranial aneurysm detection.
• This method offers a valuable tool for accurate aneurysm diagnosis.
• Enhances clinical decision-making in managing cerebrovascular diseases.