Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Towards a Reliable and Rapid Automated Grading System in Facial Palsy Patients: Facial Palsy Surgery Meets Computer Science.

Journal of clinical medicine·2022
Same author

Correction to: New aspects in digital breast assessment: further refinement of a method for automated digital anthropometry.

Archives of gynecology and obstetrics·2021
Same author

New aspects in digital breast assessment: further refinement of a method for automated digital anthropometry.

Archives of gynecology and obstetrics·2020
Same journal

Model-guided medicine for early diagnosis of transthyretin-associated cardiac amyloidosis using multimodal data integration and standardized interoperable models (the CRONOS-ATTR study).

International journal of computer assisted radiology and surgery·2026
Same journal

Electromagnetic navigation for femoral osteotomy using high-accuracy X-ray-to-CT registration.

International journal of computer assisted radiology and surgery·2026
Same journal

Modular instrument actuation unit for robotic-assisted systems in laparoscopic surgery.

International journal of computer assisted radiology and surgery·2026
Same journal

Pose-aware deep perceptual similarity for iterative 2D/3D registration of knee joints using contrastive learning.

International journal of computer assisted radiology and surgery·2026
Same journal

SurgCheck: Do vision-language models really look at images in surgical VQA?

International journal of computer assisted radiology and surgery·2026
Same journal

An evolutionary neural architecture search for magnetic resonance image reconstructions.

International journal of computer assisted radiology and surgery·2026
See all related articles
  1. Home
  3. Enhancing Generalization In Zero-shot Multi-label Endoscopic Instrument Classification.
  1. Home
  3. Enhancing Generalization In Zero-shot Multi-label Endoscopic Instrument Classification.

Related Experiment Video

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

2.7K

Enhancing generalization in zero-shot multi-label endoscopic instrument classification.

Raphaela Maerkl1, Tobias Rueckert2,3, David Rauber2

  • 1Regensburg Medical Image Computing (ReMIC), OTH Regensburg, 93053, Regensburg, Germany. raphaela.maerkl@st.oth-regensburg.de.

International Journal of Computer Assisted Radiology and Surgery
|June 11, 2025

View abstract on PubMed

Summary
This summary is machine-generated.

Improving zero-shot learning for medical AI, this study uses sentence embeddings and z-score normalization to enhance recognition of unseen surgical instruments, boosting accuracy significantly.

Keywords:
Generalized zero-shot learningMulti-label classificationSentence embeddingsSurgical instrumentsZ-score normalization

More Related Videos

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

2.4K
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

483

Related Experiment Videos

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
04:48

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

2.7K
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
05:56

Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application

Published on: April 14, 2023

2.4K
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

483

Area of Science:

  • Computer Vision
  • Machine Learning
  • Medical Imaging

Background:

  • Neural networks struggle with generalizing to unseen classes, a critical issue in safety-critical medical applications.
  • Zero-shot learning (ZSL) offers a solution by leveraging semantic data, but performance hinges on embedding quality.

Purpose of the Study:

  • To investigate the efficacy of full descriptive sentence embeddings versus simpler word embeddings for ZSL in medical image recognition.
  • To evaluate the impact of z-score normalization on embedding performance for unseen classes.

Main Methods:

  • Utilized Sentence-BERT for generating descriptive sentence embeddings as class representations.
  • Compared sentence embeddings with BERT-derived word embeddings.
  • Applied z-score normalization as a post-processing step.
  • Evaluated on a multi-label generalized zero-shot learning task for surgical instrument recognition in endoscopic images.

    • Main Results:

    • Combining sentence embeddings with z-score normalization significantly improved performance on unseen classes.
    • Area Under the Receiver Operating Characteristic Curve (AUROC) for unseen classes increased from 43.9% to 64.9%.
    • Multi-label accuracy for unseen classes rose from 26.1% to 79.5%.

    Conclusions:

    • Sentence embeddings and z-score normalization substantially enhance the generalization capabilities of zero-shot learning models.
    • The proposed method shows promise for improving reliability in medical AI applications.
    • Further validation across diverse datasets and domains is recommended to confirm robustness.