Enhanced framework for COVID-19 prediction with computed tomography scan images using dense convolutional neural...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Infectious Disease Diagnostics

Background:

• Computed Tomography (CT) scans are increasingly used for COVID-19 characterization.
• Deep learning (DL) methods, particularly Convolutional Neural Networks (CNNs), show promise for COVID-19 diagnosis.
• Inefficient classification models risk high 'False Negative' rates, endangering patient lives.

Purpose of the Study:

• To enhance COVID-19 patient classification accuracy using CT scans.
• To minimize the occurrence of 'False Negatives' where COVID-19 is misclassified as 'Non-Covid'.
• To improve the convergence and efficiency of CNN algorithms for medical diagnosis.

Main Methods:

• Utilized a Dense-CNN architecture for efficient patient categorization.
• Developed and applied a novel cross-entropy-based loss function to optimize CNN convergence.
• Trained and validated the proposed model on a large, recently published dataset.

Main Results:

• The proposed Dense-CNN model achieved a prediction accuracy of 93.78%.
• The model demonstrated a low false-negative rate of only 6.5%.
• Comparative analysis confirmed the proposed method's superior effectiveness over existing models.

Conclusions:

• The enhanced Dense-CNN model offers a more accurate and reliable method for COVID-19 diagnosis from CT scans.
• Reducing false negatives is critical for timely intervention and patient outcomes.
• This approach has the potential to significantly accelerate COVID-19 diagnosis and treatment pathways.