COVID-19 detection in lung CT slices using Brownian-butterfly-algorithm optimized lightweight deep features.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate COVID-19 detection is crucial, with lung CT scans playing a vital role in diagnosis and severity assessment.
• Deep learning methods offer automated solutions to reduce the diagnostic burden in analyzing medical images.
• Lightweight deep learning methods (LDMs) provide a pathway for less complex yet highly accurate detection systems.

Purpose of the Study:

• To develop a simple disease assessment scheme (DAS) for COVID-19 detection using lightweight deep learning methods (LDMs) applied to lung CT slices.
• To enhance COVID-19 diagnostic accuracy through efficient feature extraction and optimization.
• To evaluate the performance of the proposed DAS using individual, fused, and ensemble features.

Main Methods:

• Image acquisition and preprocessing using Shannon's thresholding.
• Deep-feature extraction via pre-trained lightweight deep learning methods (LDMs).
• Feature optimization with the Brownian Butterfly Algorithm (BBA) and binary classification using three-fold cross-validation.

Main Results:

• The proposed DAS achieved 93.80% accuracy with individual features.
• Fused features improved detection accuracy to 96%.
• Ensemble features yielded the highest accuracy at 99.10% for COVID-19 detection.

Conclusions:

• The developed lightweight deep learning-based DAS significantly enhances COVID-19 detection accuracy from lung CT scans.
• The integration of LDMs, BBA optimization, and ensemble feature strategies proves effective for accurate disease assessment.
• This approach offers a promising, simplified yet powerful tool for augmenting COVID-19 diagnosis.