Related Experiment Video
Updated: Sep 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Effective multiscale deep learning model for COVID19 segmentation tasks: A further step towards helping radiologist
Abdul Qayyum1, Alain Lalande1,2, Fabrice Meriaudeau1
1ImViA Laboratory, University of Bourgogne Franche-Comt́e, Dijon, France.
Neurocomputing
|May 17, 2022
Summary
A new deep learning model effectively segments COVID-19 lung infections in CT scans. This approach ensures generalizable performance on various datasets, aiding clinical applications and disease progression studies.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Pulmonology
Background:
- The COVID-19 pandemic necessitates accurate diagnostic and evaluation tools, with CT scans being a key area of investigation.
- Current deep learning models for lung CT segmentation suffer from poor generalization due to limited, non-standardized datasets.
- Accurate segmentation of infected lung areas in CT scans is crucial for understanding COVID-19 progression.
Purpose of the Study:
- To develop and validate a novel deep learning model for accurate COVID-19 segmentation in CT images.
- To address the limitations of existing models by employing a multiscale and multilevel feature extraction strategy.
- To assess the model's generalization capabilities across diverse datasets and segmentation tasks.
Main Methods:
- A novel deep learning model featuring a unique encoder-decoder architecture was developed.
- The model incorporates a kernel-based atrous spatial pyramid pooling module for multiscale feature extraction.
- A multistage skip connection concatenation approach was utilized to integrate features effectively.
Main Results:
- The proposed model achieved high Dice scores across multiple datasets: 90% (100 cases), 95% (NSCLC), 88.49% (COVID-19), and 97.33% (StructSeg 2019).
- Demonstrated strong generalizable performance, even when trained on smaller datasets.
- Outperformed existing state-of-the-art models in segmentation accuracy.
Conclusions:
- The developed deep learning model offers a robust and generalizable solution for COVID-19 lung segmentation.
- The model's effectiveness supports its potential application in clinical settings for disease management.
- Publicly available source code facilitates further research and development in medical image analysis.
