A multi-task generative model for simultaneous post-contrast MR image synthesis and brainstem glioma segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Gadolinium-based contrast agents (GBCAs) are essential for brainstem glioma (BSG) detection via MRI but pose potential risks.
• Reducing GBCA exposure is crucial for patient safety while maintaining diagnostic accuracy.
• Accurate segmentation of BSG lesions is vital for effective treatment planning.

Purpose of the Study:

• To develop a generative model for synthesizing post-contrast T1-weighted (T1w) MR images for BSG detection.
• To simultaneously segment BSG lesions from multi-contrast MRI inputs.
• To reduce reliance on GBCAs while enhancing diagnostic information.

Main Methods:

• A retrospective analysis of 30 BSG patients' multi-contrast MR images (pre-T1w, T2w, ASL, post-contrast T1w).
• Development of a multi-task generative model for synthesizing post-contrast T1w images and segmenting BSG masks.
• Performance evaluation using PSNR, SSIM, MAE, and Dice Similarity Coefficient (DSC), complemented by a perceptual study.

Main Results:

• The model achieved high image synthesis quality with SSIM of 0.86 ± 0.04, PSNR of 26.33 ± 0.05, and MAE of 57.20 ± 20.50.
• Automated BSG lesion segmentation yielded a satisfactory DSC score of 0.88 ± 0.27.
• Perceptual studies confirmed the diagnostic quality of synthesized images.

Conclusions:

• The proposed model effectively synthesizes high-quality post-contrast T1w MR images for BSG detection.
• The model accurately segments BSG lesions, demonstrating potential for clinical application.
• This approach offers a promising method to reduce GBCA usage and improve diagnostic efficiency.