Encoder-decoder convolutional neural network for simple CT segmentation of COVID-19 infected lungs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Manual segmentation of COVID-19 infected regions in CT scans is time-consuming and resource-intensive.
• Existing deep learning models for segmentation often involve complex architectures and extensive training requirements.
• There is a need for accessible and efficient automated segmentation tools for clinical application.

Purpose of the Study:

• To develop and evaluate a simple, reproducible Encoder-Decoder convolutional neural network (ED-CNN) model for automated segmentation of COVID-19 CT data.
• To compare the performance of the proposed ED-CNN model against existing, more complex deep learning approaches.
• To demonstrate the utility of automated segmentation in clinical workflows for faster assessment of lung infection.

Main Methods:

• Application of a compact Encoder-Decoder convolutional neural network (ED-CNN) model.
• Training and validation on COVID-19 computerised tomography (CT) datasets.
• Evaluation of segmentation performance using metrics such as Specificity, Accuracy, and Mean Absolute Error.

Main Results:

• The ED-CNN model achieved high segmentation accuracy, with Specificity (Sp) = 0.996 ± 0.001, Accuracy (Acc) = 0.994 ± 0.002, and Mean Absolute Error (MAE) = 0.0075 ± 0.0005.
• The proposed model utilizes significantly fewer parameters (approx. 49k) compared to other complex deep learning networks.
• Automated segmentation successfully delineated infected regions in thoracic CT scans, comparable to results from more complex models.

Conclusions:

• The developed ED-CNN model provides a high-quality, automated solution for segmenting COVID-19 CT scans.
• Its simplicity, efficiency, and low parameter count make it accessible for real-world applications and personalized medicine workflows.
• Automated segmentation can significantly expedite the diagnosis and treatment planning process, saving valuable time and resources.