Gigapixel Whole-Slide Images (WSIs) present a fundamental computational bottleneck for vision-language models (VLMs) due to extreme sequence lengths. Existing approaches predominantly rely on spatial sampling or training-free pruning, which risk diluting weak but informative signals, leading to the loss of critical diagnostic evidence due to the spatially diffuse nature of pathological cues. We reformulate WSI token pruning as a sequential selection process, enabling the model to autonomously learn an optimal routing strategy rather than relying on static heuristics. We herein propose a decoupled routing framework integrated as an active plugin into the fully pre-trained SlideChat base model, leaving both the slide encoder and large language model frozen. To provide continuous gradients for the non-differentiable pruning operation during training, we introduce PathSelect. PathSelect employs a variance-preserving noise gate to modulate each patch's information flow via a differentiable Soft Top-K operator, paired with a diagonal-attention Denoiser that recovers the perturbed representations without semantic leakage. At inference, the PathSelect module is entirely detached. Relying solely on the trained Scorer, a deterministic Hard Top-K operator executes adaptive, data-dependent trajectory termination, significantly accelerating downstream generative processing with exceptionally low sequential token selection latency. Driven by an empirical average of only 44.86 tokens under a maximum constraint of K = 128, our framework achieves 74.00% overall accuracy on SlideBench (TCGA), representing an approximate 36.6x spatial token reduction relative to the uncompressed baseline average while consistently outperforming sampling-based counterparts.
Jing-Zhi Chen, Lan He, Zehong Chen et al.· 1 citation
Recent visual-text compression (VTC) methods, typified by DeepSeek-OCR, report impressive high token compression ratios for long-context modeling tasks by leveraging text-to-image rendering. However, existing evaluation protocols heavily rely on downstream task performance. Such evaluation metrics fail to accurately measure text preservation due to the strong inherent linguistic priors of Multimodal Large Language Models (MLLMs). In this work, we introduce a new evaluation framework that decouples MLLMs'capabilities to faithfully assess VTC quality. Within this framework, we further introduce the ZeroSense Benchmark to ensure low semantic correlation of testing samples. By eliminating textual dependencies, our benchmark guarantees that the evaluation results are purely reflective of VTC quality, unaffected by the semantic inference capabilities of downstream models. Extensive experiments across multiple datasets demonstrate that VTC quality and downstream task accuracy diverge significantly, highlighting the necessity of our decoupled evaluation framework.
Yonghan Gao, Zehong Chen, Lijian Xu et al.· 0 citations
Introduction The superior clinical utility of 7T magnetic resonance imaging (MRI) is constrained by high acquisition costs and limited scanner availability. While deep learning-based 3T-to-7T synthesis offers a potential solution, prevailing models typically rely on heavy parameterization, which increases computational redundancy and risk of overfitting on restricted medical datasets. In this paper, we focus on model efficiency and propose LiteMamba-Synth, an architecturally streamlined state space framework designed for high-fidelity MRI translation with minimal resource requirements. Methods Our core contribution is the integration of the ConvMamba block, which utilizes the linear-time complexity of State Space Models (SSMs) to capture expansive spatial dependencies without the prohibitive computational overhead of traditional attention mechanisms. To preserve essential anatomical details during the compression of the feature space, we introduce the Wavelet-Enhanced Skip connection (WES), a module that facilitates multi-scale frequency-domain feature fusion to safeguard high-frequency textures and edge information. Additionally, a lightweight Convolutional Block Attention Module (CBAM) is incorporated to adaptively recalibrate feature responses toward salient neuroanatomical regions. Results Experimental results on the UNC T1w dataset demonstrate that LiteMamba-Synth achieves a competitive PSNR of 20.82 dB and an SSIM of 0.711. Crucially, our model maintains a compact footprint of merely 2.15 million parameters, representing a substantial reduction in complexity compared to contemporary deep learning baselines. Discussion By delivering high-quality synthesis results with minimal parameter overhead, LiteMamba-Synth provides a practical and scalable solution for deploying advanced MRI synthesis in resource-constrained clinical environments.
Zhengrui Zhang, Jie Dong, Haoting Yang et al.· Frontiers in Neuroanatomy· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
