Evaluation of EfficientNet models for COVID-19 detection using lung parenchyma.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• The COVID-19 pandemic highlighted the need for rapid and accurate diagnostic tools.
• Computer Tomography (CT) scans are vital for detecting COVID-19 pneumonia.
• Efficient diagnostic methods are crucial for controlling virus spread and reducing mortality.

Purpose of the Study:

• To develop an open-source, CT-based image dataset for COVID-19 detection.
• To evaluate the effectiveness of deep learning models for COVID-19 diagnosis using CT scans.
• To introduce and validate a modified EfficientNet model with a novel activation function.

Main Methods:

• Generation of a novel CT image dataset comprising 180 COVID-19-positive and 86 COVID-19-negative cases.
• Application of k-means clustering for smart segmentation as a preprocessing step.
• Evaluation of various pre-trained Convolutional Neural Network (CNN) architectures, including modified EfficientNet models with the Nish activation function.

Main Results:

• The modified EfficientNet-B4-ap-nish model demonstrated superior performance on the developed dataset.
• Achieved an accuracy rate of 97.93% and an F1-score of 97.33% for COVID-19 detection.
• The proposed method effectively utilizes the CT scan dataset for diagnostic purposes.

Conclusions:

• The developed open-source CT dataset and the EfficientNet-B4-ap-nish model offer a promising approach for accurate COVID-19 diagnosis.
• This research contributes significantly to improving early detection capabilities and managing the pandemic.
• The findings have substantial implications for current clinical applications and future advancements in AI-driven medical diagnostics.