[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-85302-en":3,"doc-seo-85302-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},85302,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","LaGuadia Language-Guided Adaptive Distillation from Pathology Foundation Models","Pathology Foundation Models (PFMs) provide strong Whole Slide Image (WSI) representations but require massive computation. Existing multi-teacher knowledge distillation often uses fixed or uniform weighting, overlooking tissue-level semantic and spatial heterogeneity. LaGuadia introduces language-guided adaptive distillation that builds a compact pathology image encoder by extracting clinical keywords from pathology reports, aligning visual features to them through a vision-language meta-teacher, and weighting each teacher by semantic alignment to the clinical narrative.","arXiv :2607 . 11257v1 [ cs .CV] 13 Jul 2026  \nLaGuadia: Language-Guided Adaptive Distillation from Pathology Foundation Models  \nGangsu Kim and Won-Ki Jeong†  \nDepartment of Computer Science and Engineering, College of Informatics, Korea University, Republic of Korea  \n{gangsu1813, [wkjeong}@korea.ac.kr](wkjeong}@korea.ac.kr)  \nAbstract. 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](https://github.com/hvcl/LaGuadia).  \nKeywords: Whole Slide Image · Knowledge Distillation · Clinical Language Knowledge  \n1 Introduction  \nComputational pathology has rapidly advanced with the emergence of Pathology Foundation Models (PFMs), which leverage hundreds of millions of patches extracted from large-scale Whole Slide Image (WSI) datasets. Early studies primarily focused on vision-only PFMs that learn fine-grained morphological patterns in a self-supervised manner [4, 6, 26, 29] . More recently, Vision-Language PFMshave further enhanced pathology representation learning by integrating visual information with clinical reports, either through joint embedding spaces [12] or  \n† Corresponding author: [wkjeong@korea.ac.kr](wkjeong@korea.ac.kr)  \n2 Kim et al.  \nfused architectures [25] . This integration enables richer diagnostic context beyond pure morphology, substantially improving representational expressiveness. However, the massive parameter scale of such PFMs incurs high computational costs, limiting their practicality in real-world clinical settings. Thus, developing lightweight yet expressive student models is crucial for improving the accessibility of digital pathology.  \nKnowledge distillation (KD) has emerged as an effective strategy for transferring the representational capacity of large PFMs to compact student networks. In pathology, KD methods can be broadly categorized into single-teacher [5, 27, 24] and multi-teacher approaches [14, 23, 7, 21] . While single-teacher distillation is constrained by the bias of a fixed expert, multi-teacher KD seeks to leverage complementary strengths across models. However, existing multi-teacher methods typically rely on uniform or coarse confidence-based aggregation, failing to reflect the strong spatial and semantic heterogeneity of pathology images. As a result, the optimal teacher may vary substantially across tissue regions.  \nDespite the need for adaptive teacher selection, assessing teacher expertise solely from visual features remains ambiguous due to the lack of a reliable ground truth, as visually similar tissue regions may correspond to distinct diagnostic interpretations depending on clinical context. Since pathological diagnosis is ultimately summarized","cbCairCCffxcSW9T","https://ap.wps.com/l/cbCairCCffxcSW9T","pdf",693254,3,1,11,"English","en",105,"# Introduction\n# Keyword Extraction\n## Align vision-language embedding","[{\"question\":\"What problem does LaGuadia address in knowledge distillation for pathology?\",\"answer\":\"LaGuadia addresses the high computational cost of PFMs and the limitation of existing multi-teacher distillation that uses suboptimal uniform weighting, which ignores tissue heterogeneity.\"},{\"question\":\"How does LaGuadia use clinical language during training?\",\"answer\":\"It extracts visually observable clinical keywords from pathology reports, aligns visual features with these keywords using a vision-language meta-teacher, and then adaptively weights teacher contributions based on semantic alignment with the clinical narrative.\"},{\"question\":\"What tasks and results are reported to validate LaGuadia?\",\"answer\":\"Experiments on WSI captioning, visual question answering, and slide-level classification show that an 87M-parameter LaGuadia student matches or exceeds foundation-scale models while maintaining factual consistency and robust generalization.\"}]",1784202344,28,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"laguadia-language-guided-adaptive-distillation-from-pathology-foundation-models","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/laguadia-language-guided-adaptive-distillation-from-pathology-foundation-models/85302/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does LaGuadia address in knowledge distillation for pathology?","Question",{"text":75,"@type":76},"LaGuadia addresses the high computational cost of PFMs and the limitation of existing multi-teacher distillation that uses suboptimal uniform weighting, which ignores tissue heterogeneity.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does LaGuadia use clinical language during training?",{"text":80,"@type":76},"It extracts visually observable clinical keywords from pathology reports, aligns visual features with these keywords using a vision-language meta-teacher, and then adaptively weights teacher contributions based on semantic alignment with the clinical narrative.",{"name":82,"@type":73,"acceptedAnswer":83},"What tasks and results are reported to validate LaGuadia?",{"text":84,"@type":76},"Experiments on WSI captioning, visual question answering, and slide-level classification show that an 87M-parameter LaGuadia student matches or exceeds foundation-scale models while maintaining factual consistency and robust generalization.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]