arXiv cs.CL· Ali Kerem Bozkurt, Baris Cem Bakay, Ibrahim Kulac, Cigdem Gunduz-Demir, Erkut Erdem, Aykut Erdem·· 11 小时前AI 评分25
SLICEChat：面向全切片病理语言模型的编码器内渐进式 token 剪枝
SLICEChat: Progressive In-Encoder Token Pruning for Whole-Slide Pathology Language Models
AI 导读
SLICEChat 是一个在混合 Mamba-Transformer 切片编码器内集成渐进式 token 剪枝的切片级病理 MLLM，可在多模态融合前将 gigapixel 级全切片图像压缩为紧凑表示。
正文 · 原文
Abstract:Whole-slide pathology images (WSIs) contain gigapixel-scale visual content, creating a major scalability challenge for slide-level multimodal large language models (MLLMs). Existing approaches process thousands of patch tokens and typically apply compression only after slide encoding, leaving multimodal attention computationally expensive. We introduce SLICEChat, a slide-level MLLM that integrates progressive token pruning within a hybrid Mamba--Transformer slide encoder. Mamba layers enable efficient long-range propagation, while Transformer layers preserve global interactions as the sequence is progressively shortened. Between stages, language-supervised, region-aware pruning removes spatially coherent low-utility regions under a controlled keep-rate schedule, producing compact slide representations before multimodal fusion. On SlideBench VQA, SLICEChat achieves 79.84% accuracy on TCGA and 59.09% on BCNB cohorts, outperforming prior slide-level pathology MLLMs, and achieves the highest overall WSI-Bench metrics. It also provides competitive memory usage and the inference latency among the evaluated models. These results demonstrate accurate and computationally efficient multimodal reasoning over gigapixel WSIs.
| Comments: | Accepted at WACV 2027 | Project page: this https URL | Code: this https URL |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Computation and Language (cs.CL) |
| Cite as: | arXiv:2609.24894 [cs.CV] |
| (or arXiv:2609.24894v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2609.24894 arXiv-issued DOI via DataCite |
Submission history
From: Ali Kerem Bozkurt [view email]
[v1]
Mon, 21 Sep 2026 16:59:06 UTC (20,022 KB)
[v2]
Wed, 4 Nov 2026 11:15:01 UTC (20,017 KB)
来源：arXiv cs.CL · arxiv.org