[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84373-en":3,"doc-seo-84373-105":29,"detail-sidebar-cat-0-en-105":90},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84373,13056703020460,"Valentina","https://ap-avatar.wpscdn.com/avatar/be000253dac470eee5d?_k=1778207105932848923",7,"Healthcare","CT-CLIP Representations for Multimodal Lung Cancer Survival Prediction","Accurate lung cancer prognosis is essential for treatment planning, yet deep survival models are constrained by limited curated imaging cohorts with reliable outcome labels. This study investigates whether a domain-specific vision-language foundation model, CT-CLIP, can produce transferable representations for multimodal survival prediction. Using pretreatment CT images and clinical variables from 242 diagnosed patients, the work evaluates frozen encoders, full fine-tuning, and low-rank adaptation, including modality ablations and baseline comparisons. A frozen CT-CLIP plus a lightweight survival head outperforms clinical baselines and yields meaningful high-/low-risk stratification.","arXiv :2607 .08503v 1 [ cs .CV] 9 Jul 2026  \nCT-CLIP Representations for Multimodal Lung Cancer Survival Prediction  \nSofie Allgöwer 1 , Mikael Johansson2 , Andreas Hallqvist3 , Jonas Andersson2 ,Åse Johnsson4 , Ida Häggström 1 , and Jennifer Alvén 1  \n1 Chalmers University of Technology, Gothenburg, Sweden  \n[allgower@chalmers.se](allgower@chalmers.se)  \n2 Umeå University, Umeå, Sweden  \n3 Sahlgrenska University Hospital, Gothenburg, Sweden  \n4 Sahlgrenska Academy at Gothenburg University, Gothenburg, Sweden  \nAbstract. Accurate prognosis prediction is important for treatment planning in lung cancer, but deep learning-driven survival modelling is often limited by the scarcity of curated imaging cohorts with reliable outcome data. This study evaluates whether representations from a domain-specific foundation model can be used for multimodal survival prediction in data-constrained clinical settings. We assess the foundation model CT-CLIP as a feature extractor for pretreatment computed tomography images and clinical variables from 242 diagnosed lung cancer patients. The evaluation includes adaptation strategies based on frozen encoders, full fine-tuning, and low-rank adaptation, together with modality ablations and comparisons with clinical and multimodal baselines. The results show that a frozen CT-CLIP model combined with a trainable lightweight survival head outperforms the clinical baseline and achieves comparable or improved performance relative to other multimodal approaches, and separates patients into clinically meaningful high-and low-risk groups.  \nKeywords: Multimodal Survival Analysis · Lung Cancer · CT-CLIP  \n1 Introduction  \nLung cancer accounts for the highest number of cancer-related deaths worldwide. Despite advances in imaging and treatment, the 5-year survival rate remains below 20% in most parts of the world [2] . This persistently poor prognosis underscores the need for improved risk stratification and prognostication. To personalize clinical decision-making and treatment planning, accurate survival prediction models are of importance.  \nOne of the primary challenges in developing such models is the limited availability of sufficiently large and well-curated public datasets with imaging, clinical variables, and reliable outcome data. In this work, we use a real-world dataset of 242 diagnosed lung cancer patients, consisting of pretreatment computed tomography (CT) scans and clinical variables, and investigate alternatives for lung cancer survival prediction in data-scarce settings.  \n2 F. Author et al.  \nRelated work. While traditional work performed unimodal survival prediction using methods such as Cox Proportional Hazards (CoxPH) [3] for tabular data, or DeepConvSurv [19] for images, more recent studies have explored multimodal survival prediction. Intuitively, such models may better reflect the prognostication process of medical doctors, who integrate information from multiple modalities when assessing patient outcomes. Several approaches employ Convolutional Neural Networks (CNNs) for CT image feature extraction combined with tabular data through various fusion strategies [12, 16, 18] . However, these models typically rely on manually annotated tumor regions as input.  \nMore recently, Kim et al. [9] developed a framework for 5-year overall survival prediction, using multiple instance learning to integrate information from CT, pathology images and clinical data. They employed a CNN as image feature extractor, converted tabular data into clinical notes and encoded the notes with CLIP’s [14] pretrained text encoder, which improved the 5-year AUC compared to standard tabular integration. Given the domain gap between medical and general language [1], it remains unclear whether a text encoder pretrained on medical data would better capture the semantic meaning of clinical notes.  \nXing et al. [17] presented a lung cancer survival prediction framework with modality-specific encoders: a 3D vision transformer (","cbCaiqevRwxICftg","https://ap.wps.com/l/cbCaiqevRwxICftg","pdf",631415,1,10,"English","en",105,"# Introduction\n## Motivation and data constraints\n# Related work\n## Unimodal survival prediction\n## Multimodal approaches\n# Contribution\n# Methodology\n## Survival analysis","[{\"question\":\"What problem does the study address in lung cancer survival modelling?\",\"answer\":\"The study targets the difficulty of accurate prognosis prediction when curated imaging cohorts with reliable outcome data are scarce, limiting deep learning survival modelling in clinical settings.\"},{\"question\":\"How is CT-CLIP used in the proposed method?\",\"answer\":\"CT-CLIP serves as a feature extractor for pretreatment CT images, combined with clinical variables to support multimodal survival prediction via adaptation strategies such as frozen encoders, full fine-tuning, and low-rank adaptation.\"},{\"question\":\"What adaptation strategies and comparisons are included in the evaluation?\",\"answer\":\"The evaluation compares frozen CT-CLIP representations with full fine-tuning and LoRA, and includes modality ablations plus comparisons against clinical and multimodal baselines, focusing on discrimination and risk-group separation.\"}]",1784195169,25,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"ct-clip-representations-for-multimodal-lung-cancer-survival-prediction","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/healthcare/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/ct-clip-representations-for-multimodal-lung-cancer-survival-prediction/84373/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What problem does the study address in lung cancer survival modelling?","Question",{"text":74,"@type":75},"The study targets the difficulty of accurate prognosis prediction when curated imaging cohorts with reliable outcome data are scarce, limiting deep learning survival modelling in clinical settings.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How is CT-CLIP used in the proposed method?",{"text":79,"@type":75},"CT-CLIP serves as a feature extractor for pretreatment CT images, combined with clinical variables to support multimodal survival prediction via adaptation strategies such as frozen encoders, full fine-tuning, and low-rank adaptation.",{"name":81,"@type":72,"acceptedAnswer":82},"What adaptation strategies and comparisons are included in the evaluation?",{"text":83,"@type":75},"The evaluation compares frozen CT-CLIP representations with full fine-tuning and LoRA, and includes modality ablations plus comparisons against clinical and multimodal baselines, focusing on discrimination and risk-group separation.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,117,122,127,130,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":115,"slug":116},40,"healthcare",{"id":118,"doc_module":4,"doc_module_name":45,"category_name":119,"show_sort_weight":120,"slug":121},8,"Research & Report",30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":21,"slug":132},"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]