<?xml version="1.0" encoding="utf-8"?>
<export-example>
  <doc>
    <id>8567</id>
    <completedYear/>
    <publishedYear>2025</publishedYear>
    <thesisYearAccepted/>
    <language>eng</language>
    <pageFirst>1577</pageFirst>
    <pageLast>1587</pageLast>
    <pageNumber/>
    <edition/>
    <issue/>
    <volume>20</volume>
    <type>article</type>
    <publisherName>Springer Nature</publisherName>
    <publisherPlace/>
    <creatingCorporation/>
    <contributingCorporation/>
    <belongsToBibliography>0</belongsToBibliography>
    <completedDate>2025-11-06</completedDate>
    <publishedDate>--</publishedDate>
    <thesisDateAccepted>--</thesisDateAccepted>
    <title language="eng">Enhancing generalization in zero-shot multi-label endoscopic instrument classiﬁcation</title>
    <abstract language="eng">Purpose &#13;
Recognizing previously unseen classes with neural networks is a signiﬁcant challenge due to their limited generalization capabilities. This issue is particularly critical in safety-critical domains such as medical applications, where accurate classiﬁcation is essential for reliability and patient safety. Zero-shot learning methods address this challenge by utilizing additional semantic data, with their performance relying heavily on the quality of the generated embeddings.&#13;
&#13;
Methods &#13;
This work investigates the use of full descriptive sentences, generated by a Sentence-BERT model, as class representations, compared to simpler category-based word embeddings derived from a BERT model. Additionally, the impact of z-score normalization as a post-processing step on these embeddings is explored. The proposed approach is evaluated on a multi-label generalized zero-shot learning task, focusing on the recognition of surgical instruments in endoscopic images from minimally invasive cholecystectomies.&#13;
&#13;
Results &#13;
The results demonstrate that combining sentence embeddings and z-score normalization signiﬁcantly improves model performance. For unseen classes, the AUROC improves from 43.9% to 64.9%, and the multi-label accuracy from 26.1% to 79.5%. Overall performance measured across both seen and unseen classes improves from 49.3% to 64.9% in AUROC and from 37.3% to 65.1% in multi-label accuracy, highlighting the effectiveness of our approach.&#13;
&#13;
Conclusion &#13;
These ﬁndings demonstrate that sentence embeddings and z-score normalization can substantially enhance the generalization performance of zero-shot learning models. However, as the study is based on a single dataset, future work should validate the method across diverse datasets and application domains to establish its robustness and broader applicability.</abstract>
    <parentTitle language="eng">International Journal of Computer Assisted Radiology and Surgery</parentTitle>
    <identifier type="doi">10.1007/s11548-025-03439-5</identifier>
    <identifier type="urn">urn:nbn:de:bvb:898-opus4-85674</identifier>
    <note>Corresponding author der OTH Regensburg: Raphaela Maerkl</note>
    <enrichment key="BegutachtungStatus">peer-reviewed</enrichment>
    <enrichment key="opus.source">publish</enrichment>
    <enrichment key="CorrespondingAuthor">Raphaela Maerkl</enrichment>
    <enrichment key="opus.doi.autoCreate">false</enrichment>
    <enrichment key="opus.urn.autoCreate">true</enrichment>
    <licence>Creative Commons - CC BY - Namensnennung 4.0 International</licence>
    <author>Raphaela Maerkl</author>
    <author>Tobias Rueckert</author>
    <author>David Rauber</author>
    <author>Max Gutbrod</author>
    <author>Danilo Weber Nunes</author>
    <author>Christoph Palm</author>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>Generalized zero-shot learning</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>Sentence embeddings</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>Z-score normalization</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>Multi-label classiﬁcation</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>Surgical instruments</value>
    </subject>
    <collection role="institutes" number="FakIM">Fakultät Informatik und Mathematik</collection>
    <collection role="institutes" number="RCBE">Regensburg Center of Biomedical Engineering - RCBE</collection>
    <collection role="institutes" number="RCHST">Regensburg Center of Health Sciences and Technology - RCHST</collection>
    <collection role="persons" number="palmremic">Palm, Christoph (Prof. Dr.) - ReMIC</collection>
    <collection role="oaweg" number="">Hybrid Open Access - OA-Veröffentlichung in einer Subskriptionszeitschrift/-medium</collection>
    <collection role="oaweg" number="">Corresponding author der OTH Regensburg</collection>
    <collection role="funding" number="">DEAL Springer Nature</collection>
    <collection role="institutes" number="">Labor Regensburg Medical Image Computing (ReMIC)</collection>
    <collection role="DFGFachsystematik" number="4">Naturwissenschaften</collection>
    <collection role="othforschungsschwerpunkt" number="">Gesundheit und Soziales</collection>
    <thesisPublisher>Ostbayerische Technische Hochschule Regensburg</thesisPublisher>
    <file>https://opus4.kobv.de/opus4-oth-regensburg/files/8567/Maerkl_EnhancingGeneralization2025.pdf</file>
  </doc>
</export-example>
