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

Vision01:24

Vision

59.2K
Vision is the result of light being detected and transduced into neural signals by the retina of the eye. This information is then further analyzed and interpreted by the brain. First, light enters the front of the eye and is focused by the cornea and lens onto the retina—a thin sheet of neural tissue lining the back of the eye. Because of refraction through the convex lens of the eye, images are projected onto the retina upside-down and reversed.
59.2K

You might also read

Related Articles

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

Sort by
Same author

AI/ML-Assisted Detection of <i>HMGA2</i> RNA Isoforms in Prostate Cancer Patient Tissue.

International journal of molecular sciences·2026
Same author

A Multimodal Adaptive Inter-Region Attention-Guided Network for Brain Tumor Classification.

IEEE access : practical innovations, open solutions·2025
Same author

A Hybrid Learning-Architecture for Mental Disorder Detection Using Emotion Recognition.

IEEE access : practical innovations, open solutions·2024
Same author

Multi-Stage Classification of Retinal OCT Using Multi-Scale Ensemble Deep Architecture.

Bioengineering (Basel, Switzerland)·2023
Same author

Multi-Stage Classification-Based Deep Learning for Gleason System Grading Using Histopathological Images.

Cancers·2022
Same author

Predicting the Level of Respiratory Support in COVID-19 Patients Using Machine Learning.

Bioengineering (Basel, Switzerland)·2022

Related Experiment Video

Updated: Jan 7, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

983

An Empirical Evaluation of Low-Rank Adapted Vision-Language Models for Radiology Image Captioning.

Mahmudul Hoque1, Raisa Nusrat Chowdhury1, Md Rakibul Hasan2

  • 1Department of Computer Science, Morgan State University, Baltimore, MD 21251, USA.

Bioengineering (Basel, Switzerland)
|December 30, 2025
PubMed
Summary

Large vision-language models (VLMs) generally outperform smaller ones in medical imaging tasks. However, targeted adaptation strategies and architectural design allow some compact models to achieve competitive performance, aiding radiologist workloads.

Keywords:
Low-Rank AdaptationROCOv2 datasetcaption qualitymedical image captioningparameter-efficient fine-tuningradiology report generationvision-language models

More Related Videos

Author Spotlight: Insights into Visual Cortex Research Through Wide-View fMRI Mapping
07:11

Author Spotlight: Insights into Visual Cortex Research Through Wide-View fMRI Mapping

Published on: December 8, 2023

2.3K

Related Experiment Videos

Last Updated: Jan 7, 2026

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

983
Author Spotlight: Insights into Visual Cortex Research Through Wide-View fMRI Mapping
07:11

Author Spotlight: Insights into Visual Cortex Research Through Wide-View fMRI Mapping

Published on: December 8, 2023

2.3K

Area of Science:

  • Artificial Intelligence in Medical Imaging
  • Computer Vision
  • Natural Language Processing

Background:

  • Increasing medical imaging volumes necessitate automated tools to support radiologists and reduce reporting delays.
  • Vision-language models (VLMs) show potential for accelerating medical report drafting through automated caption generation.
  • Systematic evaluation of VLMs with varying parameter scales is crucial for assessing clinical utility in radiology.

Purpose of the Study:

  • To evaluate and compare the performance of ten multimodal vision-language models (VLMs) fine-tuned on a large medical imaging dataset (ROCOv2).
  • To assess the impact of model scale (Large VLMs vs. Small VLMs) and adaptation strategies (Low-Rank Adaptation) on VLM performance in medical imaging.
  • To establish quantitative benchmarks for selecting VLMs for clinical applications in medical imaging interpretation.

Main Methods:

  • Ten multimodal models, including Large VLMs (LLaVA, IDEFICS-9B), Small VLMs (MoonDream2, Qwen, SmolVLM), and baseline architectures (VisionGPT2, CNN-Transformer), were fine-tuned on the ROCOv2 dataset (116,635 images, 8 modalities).
  • Low-Rank Adaptation (LoRA) was applied, focusing on optimizing performance with minimal parameter updates (<1% of total parameters).
  • Models were evaluated using relevance (semantic similarity) and factuality (concept-level correctness) metrics.

Main Results:

  • Performance stratified clearly by model scale: Large VLMs (0.273-0.317), Small VLMs (0.188-0.279), and baselines (0.154-0.177).
  • LLaVA-Mistral-7B achieved the highest overall performance, significantly outperforming the VisionGPT2 baseline.
  • MoonDream2 (a Small VLM) demonstrated competitive relevance scores, approaching the performance of some Large VLMs. Prepending modality labels showed variable effects on Small VLM performance.

Conclusions:

  • Model scale is a significant factor in VLM performance for medical imaging, with Large VLMs generally excelling.
  • Targeted adaptation strategies, such as optimized LoRA, and specific architectural designs enable compact Small VLMs to achieve competitive results.
  • These findings provide essential benchmarks for VLM selection, highlighting the potential of efficient, smaller models in clinical radiology workflows.