Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
U-KAN Makes Strong Backbone for Medical Image Segmentation and Generation
Authors: Chenxin Li, Xinyu Liu, Wuyang Li, Cheng Wang, Hengyu Liu, Yifan Liu, Zhen Chen, Yixuan Yuan
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Rigorous medical image segmentation benchmarks verify the superiority of UKAN by higher accuracy even with less computation cost. We further delved into the potential of U-KAN as an alternative U-Net noise predictor in diffusion models, demonstrating its applicability in generating task-oriented model architectures. ... Empirical evaluations on medical segmentation benchmarks show its superior performance, surpassing existing U-Net backbones with higher accuracy and lower computation cost. We also explore its potential as a U-Net noise predictor in diffusion models. |
| Researcher Affiliation | Academia | The Chinese University of Hong Kong EMAIL |
| Pseudocode | No | The paper describes the model architecture and equations (e.g., Eq. 1, 2, 3, 5, 6, 7, 8, 9, 10, 11, 12) in detail, and visualizes the pipeline in Fig. 1, but it does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/CUHK-AIM-Group/U-KAN |
| Open Datasets | Yes | BUSI. The dataset (Al-Dhabyani et al. 2020) comprises ultrasound images of normal, benign, and malignant breast cancer cases with corresponding segmentation maps. ... Gla S. The dataset (Valanarasu et al. 2021) consists of 612 Standard Definition (SD) frames (384 288) from 31 sequences... CVC-Clinic DB. The dataset (Bernal et al. 2015) is a public resource for polyp diagnosis in colonoscopy videos. |
| Dataset Splits | Yes | Segmentation U-KAN. ... Datasets were randomly split into 80% training and 20% validation subsets. |
| Hardware Specification | Yes | Segmentation U-KAN. We implemented U-KAN using Py Torch on an NVIDIA RTX 4090 GPU. |
| Software Dependencies | No | We implemented U-KAN using Py Torch on an NVIDIA RTX 4090 GPU. The paper mentions 'Py Torch' but does not specify a version number, nor does it list any other software dependencies with version numbers. |
| Experiment Setup | Yes | Segmentation U-KAN. We implemented U-KAN using Py Torch on an NVIDIA RTX 4090 GPU. For all datasets, we used a batch size of 8 and an initial learning rate of 1e-4. We employed the Adam optimizer with a cosine annealing learning rate scheduler (minimum 1e-5). The loss function combined binary cross-entropy (BCE) and dice loss. Datasets were randomly split into 80% training and 20% validation subsets. Results are reported over three random runs. We applied vanilla data augmentations (random rotation and flipping) and trained for 400 epochs. ... Diffusion U-KAN. For unconditional generation, images were cropped and resized to 64 64. All methods were benchmarked using the same training settings: 1e-4 learning rate, 1000 epochs, Adam optimizer, and cosine annealing learning rate scheduler. |