An Encoder-Decoder-Based Method for Segmentation of COVID-19 Lung Infection in CT Images

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computer Vision for Diagnostics

Background:

• The COVID-19 pandemic highlighted the need for rapid and accurate diagnostic tools.
• Computed Tomography (CT) scans are crucial for detecting lung infections, but image quality and interpretation can be challenging.
• Existing methods for lung infection detection using medical images have limitations in accuracy and effectiveness.

Purpose of the Study:

• To propose a novel multi-task deep-learning method for accurate lung infection segmentation in CT scan images.
• To address the challenge of limited labeled data in medical image analysis.
• To improve the accuracy and efficiency of diagnosing lung infections.

Main Methods:

• Developed a multi-task deep-learning model for segmenting lung regions and subsequent infection identification.
• Employed a two-stream input approach for multi-class segmentation, enabling learning from diverse features.
• Utilized multi-task learning to mitigate the impact of insufficient labeled training data.

Main Results:

• The proposed method achieved high performance in segmenting lung infections, demonstrating effectiveness even with limited data.
• Achieved a Dice similarity score of 78.6%, Sensitivity of 71.1%, Specificity of 99.3%, Precision of 85.6%, and Mean Absolute Error (MAE) of 0.062.
• Outperformed state-of-the-art methods in lung infection segmentation accuracy.

Conclusions:

• The multi-task deep-learning approach is highly effective for lung infection segmentation from CT scans.
• The method shows promise in overcoming data scarcity issues in medical AI.
• This AI-driven technique offers a significant advancement in the accurate and efficient diagnosis of lung infections.