Intracranial aneurysm segmentation with nnU-net: utilizing loss functions and automated vessel extraction

Area of Science:

• Medical Imaging
• Artificial Intelligence

Background:

• Intracranial aneurysms present diagnostic and treatment challenges.
• Accurate segmentation is crucial for clinical management.

Purpose of the Study:

• To develop a novel approach for segmenting intracranial aneurysms using 3D TOF-MRA images.
• To enhance segmentation accuracy and efficiency through hybrid loss functions and vessel information.

Main Methods:

• Utilized the no-new-U-net framework with 3D TOF-MRA images.
• Employed hybrid loss functions combining distribution-based and region-based losses.
• Incorporated automatic vessel segmentation for region of interest enhancement.

Main Results:

• The best-performing loss function significantly improved Dice coefficients (0.72 on RENJI, 0.54 on ADAM) and Sensitivity (0.69 on RENJI, 0.53 on ADAM).
• Demonstrated effectiveness on both Aneurysm Detection And SegMentation (ADAM) and Renji Hospital (RENJI) datasets.

Conclusions:

• The proposed method offers a promising solution for accurate intracranial aneurysm segmentation.
• Integration of hybrid loss functions and vessel information enhances delineation of complex aneurysm structures.
• Contributes to improved diagnosis and treatment planning for patients with intracranial aneurysms.