Multi-phase renal tumor segmentation with Slide-SAM assisted network

Ruiyang Jin, Quan Quan, S Kevin Zhou

Keywords: Segmentation, Pretrained model.

TL;DR: We propose SSA, a Slide-SAM assisted framework that improves renal tumor segmentation on non-contrast CT using prompts from contrast-enhanced phases.

Abstract: Non-contrast CT(Computed Tomography) scans often suffer from low tissue contrast and indistinct tumor boundaries, making accurate segmentation challenging. To address this, we propose SSA (Slide-SAM assisted network), a segmentation framework that leverages the pretrained Slide-SAM model guided by box prompts from contrast-enhanced CT. By transferring spatial priors across phases, SSA significantly improves segmentation accuracy on plain-phase images and achieves additional gains on enhanced phases. Experimental results highlight the effectiveness of combining vision foundation models with inter-phase guidance for robust medical image segmentation.

Submission Number: 49