Related Experiment Video
Updated: Jun 22, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Improving Surgical Scene Semantic Segmentation through a Deep Learning Architecture with Attention to Class Imbalance
Claudio Urrea1, Yainet Garcia-Garcia1, John Kern1
1Electrical Engineering Department, Faculty of Engineering, University of Santiago of Chile, Las Sophoras 165, Estación Central, Santiago 9170020, Chile.
Biomedicines
|June 27, 2024
Summary
This study enhances semantic segmentation for laparoscopic surgery images, particularly for rare structures. Optimized deep learning parameters improve accuracy and show promising results for critical surgical element identification.
Area of Science:
- Medical Imaging
- Computer Vision
- Surgical Technology
Background:
- Accurate semantic segmentation of laparoscopic surgery images is crucial for surgical guidance and analysis.
- Identifying structures with limited observations presents a significant challenge in surgical scene understanding.
Purpose of the Study:
- To develop and evaluate deep neural network architectures for robust semantic segmentation of laparoscopic surgical images.
- To specifically address the challenge of segmenting anatomical structures with low observation frequency.
Main Methods:
- Implementation and comparison of U-Net5ed, SegNet-VGG19, and DeepLabv3+ architectures.
- Experimentation with Rectified Linear Unit (ReLU), Gaussian Error Linear Unit (GELU), and Swish activation functions.
- Evaluation of Cross Entropy (CE), Focal Loss (FL), Tversky Loss (TL), Dice Loss (DiL), Cross Entropy Dice Loss (CEDL), and Cross Entropy Tversky Loss (CETL) loss functions, alongside Stochastic Gradient Descent with momentum (SGDM) and Adaptive Moment Estimation (Adam) optimizers.
Main Results:
- DeepLabv3+ and U-Net5ed architectures demonstrated superior performance.
- DeepLabv3+ with ResNet-50 backbone, Swish activation, and CETL loss achieved Mean Accuracy (MAcc) of 0.976 and Mean Intersection over Union (MIoU) of 0.977.
- The proposed parameters significantly improved semantic segmentation in the YOLOv9 architecture, outperforming its original configuration.
Conclusions:
- Optimized deep learning parameters enable robust semantic segmentation of all structures in surgical scenes, including rare ones.
- The findings offer highly competitive and promising results for semantic segmentation in laparoscopic surgery.
- Validated parameters show potential for enhancing various computer vision models in surgical applications.
