[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82886-en":3,"doc-seo-82886-105":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},82886,8796095462418,"Noah","https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780",8,"Research & Report","Comparison of Loss Functions for Robust Deep Learning-based Echocardiography Segmentation with Partially Labelled Data","Robust echocardiography image segmentation enables reliable estimation of cardiac biomarkers such as left ventricular ejection fraction, left atrial volume, and left ventricular myocardial mass. Manual approaches are time-consuming and suffer from inter- and intra-observer variability, while deep learning must remain stable despite low signal-to-noise, weak contrast, and scanner-specific artefacts. This work addresses partial label variation across multiple domains by comparing loss-based solutions for training with partially labelled data, providing recommendations for future development.","Comparison of Loss Functions for Robust Deep Learning-based Echocardiography Segmentation when Learning with Partially Labelled Data from Multiple Domains  \nIman Islam 1 , Bram Ruijsink 1 , Esther Puyol-Antn 1 , Andrew J. Reader 1 , Andrew P. King 1  \n1 School of Biomedical Engineering & Imaging Sciences, King’s College London, UK  \nof each loss function depending on the function to optimize model performance.  \nThis study represents the first investigation of techniques for handling partially-labelled data from multiple different  \na comprehensive comparison of loss-based solutions.  \nPartial Labels  \n©2026 I. Islam, B.  \nfirst imaging ex-  \namination carried out when assessing cardiac function and can provide insight into the function of the heart chambers and valves. Useful biomarkers can be estimated by segmenting echo images and computing volumes of the cardiac structures. A common biomarker that is extracted is the left ventricular (LV) ejection fraction (EF), which can be calculated from the LV end-diastolic volume (EDV) and end-systolic volume (ESV) . LV EF is routinely used for diagnosing and characterising heart failure (Bozkurt et al. , 2021) . Other useful biomarkers include the left atrium (LA) volume, which can be a robust predictor of diastolic heart failure and is often analysed for patients with atrial fibrillation (Taniguchi et al., 2019), and LV myocardial (LVM) mass, which can be used to detect cardiomyopathy (Berman  \nRuijsink, E. Puyol-Antn, A. J. Reader and A. P. King. License: CC-BY 4 .0  \net al. , 2019) .  \nThe analysis of echo examinations to estimate these biomarkers has traditionally been performed in a manual or semi-automatic manner . However, echo images are challenging to segment due to the relatively low signal-tonoise ratio and the poor contrast between the blood pool and myocardium. Furthermore, echo images can suffer from artefacts such as intensity inhomogeneities as well as differences in image characteristics across patients and scanners . Therefore, manual segmentation can be very time-consuming and is subject to intra-and inter-observer variability, which results in the biomarkers being poorly standardised across different annotators (Armstrong et al. , 2015) .  \nDeep learning models have been proposed to automate the segmentation of echo images (Leclerc et al., 2019;  \nMadani et al. , 2018; Ghorbani et al. , 2020; Puyol-Antnet al., 2022) and a number of commercial solutions now exist, such as GE Healthcare’s Vivid Cardiovascular Ultrasound and Ultromics’ EchoGo. However, for such tools tobe more widely used in clinics across the world, which will have significant variations in terms of imaging equipment, operator expertise and patient populations, it is important that the models are robust to the variations in image characteristics that will be encountered. In other applications, such as cardiac magnetic resonance, segmentation models have been trained using diverse data sources and shown to be robust to such variations (Mariscal-Harana et al. , 2023) . In this work we aim to address the methodological challenges involved in training a similarly robust model for echo segmentation.  \nOne significant challenge that must be overcome when training models to segment multiple structures with diverse datasets is the variation in label presence in the training data. For instance, a number of public datasets exist for training echo segmentation models but the manually defined labels present are different. A summary of the most commonly used datasets can be found in Table 1 with a sample from each dataset shown in Figure 1 . This variation in label presence means that the combination of the datasets will be partially-labelled, i.e. not all structures will be labelled in all training samples. Training naively with partially-labelled datasets such as this can cause a conflict in the supervision due to the same structures being labelled as foreground in some samples and background in others. This p","cbCaihCBaLcPMauC","https://ap.wps.com/l/cbCaihCBaLcPMauC","pdf",2028313,4,1,17,"English","en",105,"# Introduction\n# Related Works\n## Methods for partially-labelled segmentation","[{\"question\":\"Why is robust echocardiography segmentation important clinically?\",\"answer\":\"It supports automated estimation of cardiac biomarkers such as LV ejection fraction, LA volume, and LVM mass, which are used to diagnose and characterize heart conditions including heart failure and cardiomyopathy.\"},{\"question\":\"What makes echo image segmentation challenging for deep learning?\",\"answer\":\"Echo images have relatively low signal-to-noise ratio, poor contrast between blood pool and myocardium, and artefacts like intensity inhomogeneities, plus systematic differences across patients and scanners.\"},{\"question\":\"What is the key training challenge addressed in this study?\",\"answer\":\"Partially-labelled datasets arise when different structures are labeled in different samples, creating supervision conflicts when the same structure is treated as foreground in some data and background in others.\"}]",1784183670,43,{"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},"comparison-of-loss-functions-for-robust-deep-learning-based-echocardiography-segmentation-with-partially-labelled-data","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"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":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/comparison-of-loss-functions-for-robust-deep-learning-based-echocardiography-segmentation-with-partially-labelled-data/82886/",{"url":52,"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","2026-07-22","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},"Why is robust echocardiography segmentation important clinically?","Question",{"text":75,"@type":76},"It supports automated estimation of cardiac biomarkers such as LV ejection fraction, LA volume, and LVM mass, which are used to diagnose and characterize heart conditions including heart failure and cardiomyopathy.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What makes echo image segmentation challenging for deep learning?",{"text":80,"@type":76},"Echo images have relatively low signal-to-noise ratio, poor contrast between blood pool and myocardium, and artefacts like intensity inhomogeneities, plus systematic differences across patients and scanners.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the key training challenge addressed in this study?",{"text":84,"@type":76},"Partially-labelled datasets arise when different structures are labeled in different samples, creating supervision conflicts when the same structure is treated as foreground in some data and background in others.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"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":20,"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"]