Two-stage deep learning model with Segformer and Mix Vision Transformer for GIST segmentation on fused CT-PET

Proposed DIF-Net + Segformer-MViT model improved GIST segmentation on fused CT-PET over existing methods, though no specific numbers were reported.

• The model first fuses CT and PET images using DIF-Net, then segments the tumor with a Segformer architecture incorporating a Mix Vision Transformer encoder.
• Evaluation included segmentation metrics (IoU, Dice, pixel accuracy) and fusion quality metrics (PSNR, SSIM, MI, UIQI, edge preservation).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.