A deep learning segmentation-classification pipeline for X-ray-based COVID-19 diagnosis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Deep learning models for COVID-19 diagnosis using chest X-rays are prevalent.
• Many existing models lack segmentation modules, hindering clinical deployability.
• Differentiating COVID-19 from other lung diseases based on visual features is challenging.

Purpose of the Study:

• To develop and evaluate a deep learning pipeline incorporating a segmentation module and ensemble classifier for improved COVID-19 diagnosis.
• To address limitations in current AI models for medical image analysis in radiology.
• To enhance diagnostic accuracy for suspected COVID-19 cases.

Main Methods:

• A deep learning pipeline was designed, featuring a segmentation module and an ensemble classifier.
• The model was trained and evaluated on public chest X-ray datasets.
• Comparative analysis was performed against other similar models in the literature.

Main Results:

• The best performing model achieved an accuracy of 91%.
• The model demonstrated a sensitivity of 92% in diagnosing COVID-19.
• The study identified and discussed limitations in widely circulated public datasets.

Conclusions:

• The proposed deep learning pipeline with segmentation enhances COVID-19 diagnosis from chest X-rays.
• The inclusion of a segmentation module is crucial for deployable clinical AI models in radiology.
• The findings highlight the importance of robust model design and dataset evaluation for accurate disease diagnosis.