Sliding transformer with uncertainty estimation for vestibular schwannoma automatic segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurosurgery

Background:

• Automated segmentation of vestibular schwannoma (VS) using MRI aids clinical efficiency.
• Current methods struggle with ambiguous tumor borders and cystic regions, and lack uncertainty indication.
• Clinical translation requires segmentation results with confidence measures.

Purpose of the Study:

• To develop an automated VS segmentation method that provides both segmentation masks and uncertainty maps.
• To improve the accuracy and reliability of VS segmentation for clinical applications.

Main Methods:

• A U-shaped cascade transformer with a sliding window approach was employed.
• The model incorporates multiple sliding samples, a segmentation head, and an uncertainty head.
• Multimodal MRI data from 60 clinical VS patients were utilized.

Main Results:

• The proposed method achieved a Dice Similarity Coefficient (DSC) of 96.08% ± 1.30 on a clinical VS dataset.
• On a public VS dataset, the method achieved a mean DSC of 94.23% ± 2.53.
• Results were comparable or superior to existing state-of-the-art brain tumor segmentation methods.

Conclusions:

• The developed method efficiently segments VS and provides crucial uncertainty information.
• The uncertainty map assists clinical experts in reviewing segmentation results.
• This approach facilitates the transformation of automated segmentation into a clinical aid diagnostic tool.