Related Experiment Video
Updated: Dec 13, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
Improving Temporal Stability and Accuracy for Endoscopic Video Tissue Classification Using Recurrent Neural Networks.
Tim Boers1, Joost van der Putten1, Maarten Struyvenberg2
1Eindhoven University of Technology, Groene Loper 3, 5612 AE Eindhoven, The Netherlands.
Sensors (Basel, Switzerland)
|July 30, 2020
Summary
This study introduces a novel video-based deep learning approach for detecting early esophageal cancer. Recurrent Neural Networks, particularly LSTM, significantly improve diagnostic accuracy by analyzing temporal data in endoscopic videos.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Gastroenterology
Background:
- Early Barrett's neoplasia detection is challenging due to subtle visual cues and endoscopist inexperience.
- Automated detection in still endoscopic images shows promise, but video-based temporal analysis remains underdeveloped.
Discussion:
- This research leverages the temporal stability of endoscopic video data to develop a robust tissue classification framework.
- Recurrent Neural Network (RNN) nodes, specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), are integrated with a Resnet18 feature extractor.
- The study compares LSTM and GRU against Fully Connected (FC) and averaged FC classifiers.
Key Insights:
- The LSTM classifier achieved the highest accuracy (85.9%) in classifying esophageal tissue from pullback videos, outperforming FC, FC Avg(n=5), and GRU classifiers.
- The developed framework utilizes spatio-temporal information, enhancing the robustness and accuracy of endoscopic tissue classification.
- This represents a significant advancement towards temporal learning for esophageal cancer detection in endoscopic video.
Outlook:
- Further development of spatio-temporal models can improve early cancer detection rates.
- This approach could lead to more reliable diagnostic tools for endoscopists.
- The framework provides a foundation for future research in real-time endoscopic video analysis for cancer diagnosis.
