Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: Jan 7, 2026

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

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

Published on: November 30, 2022

Deep learning for deep learning performance: How much data is needed for segmentation in biomedical imaging?

Junhyeok Lee¹, Hyungjin Chung², Minseok Suh³

¹Interdisciplinary Program in Cancer Biology, Seoul National University College of Medicine, Seoul, Republic of Korea.

|December 31, 2025

Summary

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

Adverse Event Monitoring with Imaging: Prognostic Significance in Atezolizumab plus Bevacizumab Therapy for Unresectable Hepatocellular Carcinoma.

Liver cancer·2026

Same author

Feasibility of [<sup>99m</sup>Tc]Tc-Mannosylated Human Serum Albumin for Lymphoscintigraphy: A Phase 1/2 Clinical Trial.

Nuclear medicine and molecular imaging·2026

Same author

Deep Learning for Survival Prediction in Glioblastoma: Time-dependent Model Interpretability Using MRI, Clinical, and Molecular Data.

Radiology. Artificial intelligence·2026

Same author

Efficacy and Safety of Switching from Entecavir to Tenofovir Alafenamide in Chronic Hepatitis B: A Multicenter Randomized Trial in Korea.

Gut and liver·2026

Same author

Impact of Thyroid CT on Detecting Macroscopic Nodal Metastasis in Patients With Papillary Thyroid Microcarcinoma.

Korean journal of radiology·2026

Same author

Feasibility study of streamlining radioembolization with yttrium-90 resin microspheres without lung shunt fraction measurement for colorectal liver metastasis < 7 cm.

Japanese journal of radiology·2026

Same journal

Invaders taking over-Mollusc faunal change in volcanic barrier lakes of the Albertine Rift biodiversity hotspot.

PloS one·2026

Same journal

AI-driven molecular diversification and ligand-based optimization of macitentan derivatives targeting VEGFR1 and endothelin signaling pathways.

PloS one·2026

Same journal

Performance patterns and records in the world aquatics masters championships: Where do the most frequently represented nations among the top-ten masters swimmers come from?

PloS one·2026

Same journal

Modeling diurnal Temperature-Rainfall relationships under multicollinearity using PLS-SEM: A case study of Ghana.

PloS one·2026

Same journal

Organizational culture, social capital, and emergency capacity in primary healthcare institutions: A cross-sectional structural equation modeling study comparing ordinary and older communities.

PloS one·2026

Same journal

Impact of kidney function on the metabolome in the general population.

PloS one·2026

See all related articles

This summary is machine-generated.

Estimating dataset size for deep learning (DL) in medical imaging is crucial. This study introduces a framework using Long Short-Term Memory (LSTM) networks to predict segmentation performance, showing moderate data is often sufficient for viable DL models.

Area of Science:

Biomedical Imaging
Artificial Intelligence
Medical Image Analysis

Background:

Deep learning (DL) models are vital in biomedical imaging for quantitative analysis via image segmentation.
Traditional sample size estimation methods fail for DL due to high-dimensional data and nonlinear learning.
Accurate dataset size estimation is critical for efficient DL model development in clinical settings.

Purpose of the Study:

To propose a DL-specific framework for estimating the minimal dataset size required for stable segmentation performance.
To validate the framework on colorectal polyp and glioma segmentation tasks.
To introduce a surrogate modeling pipeline using Long Short-Term Memory (LSTM) networks for performance prediction.

Main Methods:

Trained residual U-Nets on varying data subsets (2%-100% for 2D, 5%-100% for 3D) for colorectal polyp and glioma segmentation.

Related Experiment Videos

Last Updated: Jan 7, 2026

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

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

Published on: November 30, 2022

Analyzed performance metrics like Dice Similarity Coefficient (DSC) to identify data saturation points.

Developed a surrogate modeling pipeline using uni-directional LSTM networks to predict segmentation performance curves.

Main Results:

Segmentation performance (DSC) improved with data and model depth, plateauing around 80% data usage in both tasks.
Best configurations achieved DSC of 0.86 for polyps and 0.79 for gliomas.
LSTM models accurately forecasted final DSC with low mean absolute error, demonstrating reliable performance prediction.

Conclusions:

Segmentation performance can be reliably estimated using lightweight surrogate models like LSTMs.
Collecting a moderate amount of high-quality data is often sufficient for developing clinically viable DL models.
The proposed framework offers a practical method for optimizing resource allocation in medical AI development.