Feasibility Study of Automated Segmentation Techniques for Pituitary Adenomas

Overview

This study evaluates the feasibility of automated segmentation techniques for pituitary adenomas, pituitary gland, and internal carotid arteries using contrast-enhanced T1-weighted MRI. Two deep learning models, Swin-Unet and nnU-Net, were assessed for their performance in segmentation accuracy and processing time.

Background

Pituitary adenomas are common intracranial neoplasms that require precise imaging for surgical planning and risk assessment. Current MRI protocols exhibit variability, which can hinder the application of automated segmentation methods in clinical practice.

Data Highlights

Key Findings

• Both Swin-Unet and nnU-Net demonstrated comparable voxel-wise performance for segmentation.
• Mean Dice similarity coefficient (DSC) was 0.70 for Swin-Unet and 0.69 for nnU-Net.
• Segmentation accuracy was lower for the pituitary gland compared to pituitary adenomas and internal carotid arteries.
• Mean Hausdorff distance (HD95) was 5.21 mm for Swin-Unet and 5.30 mm for nnU-Net.
• Mean F1-score for slice-level recognition was 0.89 for Swin-Unet and 0.87 for nnU-Net.
• Mean processing time per patient was 78.4 seconds.

Clinical Implications

The findings suggest that while voxel-level accuracy is modest, the rapid processing and reliable slice-level recognition could enhance preoperative evaluations.

Conclusion

This study supports the feasibility of using automated segmentation methods for pituitary adenomas and related structures in clinical settings.

Related Resources & Content

1. European Radiology, 2023 -- Enhanced Diagnostic Accuracy for Pituitary Microadenomas in Cushing’s Syndrome Using High-Resolution 3D Fast Spin Echo MRI
2. Int. Journal of Computer Assisted Radiology and Surgery, 2026 -- A benchmark of methods for surgical instrument segmentation in endoscopic pituitary surgery
3. European Radiology, 2023 -- Rapid and Reliable Brain Extraction from Contrast-Enhanced T1-Weighted MRI in Tumor Presence: An Enhanced Model Utilizing Multi-Center Data
4. Comparison of Manual and Semi-Automated Techniques for Measuring Vestibular Schwannoma Volume via MRI
5. Pituitary incidentaloma: a Pituitary Society international consensus guideline statement | Nature Reviews Endocrinology
6. Pituitary incidentaloma: a Pituitary Society international consensus guideline statement | Nature Reviews Endocrinology
7. Checking your browser - reCAPTCHA
8. Systematic review of pituitary gland and pituitary adenoma automatic segmentation techniques in magnetic resonance imaging