Search papers, labs, and topics across Lattice.
This paper introduces KANResDiff, a novel approach for ambiguous medical image segmentation that leverages a Kolmogorov-Arnold Network to learn local residual diffusion. By utilizing Independent Time Encoding for spline-based time embeddings and a Residual Schrodinger Bridge for flexible deterministic-stochastic interactions, the method enhances semantic modeling across inference stages. Experimental results show that KANResDiff achieves state-of-the-art performance on GED and HM-IoU metrics, with improvements of up to 16.8% and 7.7%, respectively, while remaining competitive on the MDM metric.
KANResDiff significantly boosts ambiguous medical image segmentation performance by up to 16.8% through innovative stage-aware modeling techniques.
Ambiguous medical image segmentation aims to provide a series of diverse but plausible segmentation hypotheses. However, existing methods introduce stochasticity in a fixed and pre-defined manner, failing to form a progressive semantic modeling process. To address these challenges, we propose KANResDiff to learn local residual diffusion with Kolmogorov-Arnold Network, thereby assigning distinct roles across stages for ambiguity modeling. Specifically, we propose Independent Time Encoding that offers spline-based time embeddings instead of linear ones from MLPs, which enhances the independence across inference stages and assigns progressive semantic roles to different stages. We propose Residual Schrodinger Bridge that injects deterministic residual prior with learnable weights by constructing local Schrodinger Bridge instead of following manually settings, achieving a flexible deterministic-stochastic interaction and stage-aware ambiguity modeling thanks to local optimal diffusion path. Extensive experimental results on two public datasets demonstrate that KANResDiff achieves SOTA performance on GED and HM-IoU, with maximum improvements of 16.8% and 7.7%, respectively, while maintaining competitive performance on the MDM metric. Source code is available at https://github.com/PerceptionComputingLab/KANResDiff.
