Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces LaGuadia, a novel framework that enhances the efficiency of Pathology Foundation Models (PFMs) by employing language-guided adaptive distillation. By leveraging clinical linguistic cues to dynamically weight contributions from multiple PFMs, LaGuadia creates a compact image encoder that matches or surpasses the performance of larger foundation models in various tasks such as WSI captioning and visual question answering. The results demonstrate that integrating clinical language significantly improves both factual consistency and generalization in digital pathology applications.
Clinical language can effectively guide the distillation process, enabling a compact model to rival larger foundation models in pathology tasks.
Pathology Foundation Models (PFMs) offer powerful Whole Slide Image (WSI) representations but suffer from massive computational costs. While Knowledge Distillation (KD) can create efficient student models, existing multi-teacher methods often use suboptimal uniform weighting that ignores tissue heterogeneity. We propose LaGuadia (Language-Guided Adaptive DistillAtion), a framework that develops a compact pathology image encoder by dynamically integrating expertise from multiple PFMs under clinical linguistic guidance. Our approach utilizes a multi-stage pipeline: first, extracting visually observable clinical keywords from pathology reports; second, aligning visual features with these keywords via a Vision-Language meta-teacher (MedSigLIP) to provide dense semantic guidance; and finally, performing adaptive KD where teacher contributions are weighted based on their semantic alignment with the clinical narrative. Experiments on WSI captioning, visual question answering, and slide-level classification tasks demonstrate that an 87M parameter LaGuadia student model matches or exceeds foundation-scale models such as GigaPath and UNI, achieving strong factual consistency and robust generalization. These results highlight clinical language as an effective semantic anchor for building efficient and reliable digital pathology systems. Code is available at https://github.com/hvcl/LaGuadia.