A wavelet-based deep learning pipeline for efficient COVID-19 diagnosis via CT slices

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate and rapid diagnosis of COVID-19 is crucial for disease control.
• Computed tomography (CT) scans are effective but complex to analyze, potentially delaying diagnosis.
• Deep learning (DL) offers potential for automating and accelerating COVID-19 diagnosis from CT scans.

Purpose of the Study:

• To propose CoviWavNet, a DL pipeline for automatic COVID-19 diagnosis from 3D multiview CT scans.
• To evaluate the effectiveness of spectral-temporal information compared to spatial information for diagnosis.
• To investigate the combined use of spectral-temporal and spatial features for enhanced diagnostic accuracy.

Main Methods:

• Utilized a 3D multiview dataset (OMNIAHCOV) and a public benchmark (SARS-COV-2-CT-Scan).
• Applied multilevel discrete wavelet decomposition (DWT) to CT slices, extracting spectral-temporal features from heatmaps.
• Trained ResNet CNN models using spectral-temporal information and integrated these features with spatial features for SVM classification.

Main Results:

• DL models trained on spectral-temporal information from DWT heatmaps outperformed those using only spatial information from original CT images.
• Integrating spectral-temporal and spatial features significantly improved classification accuracy.
• CoviWavNet achieved high diagnostic accuracies of 99.33% and 99.7% on the OMNIAHCOV and SARS-COV-2-CT-Scan datasets, respectively.

Conclusions:

• CoviWavNet demonstrates superior performance in COVID-19 diagnosis compared to existing methods.
• The pipeline effectively leverages spectral-temporal information for enhanced diagnostic accuracy.
• CoviWavNet can serve as a valuable tool for radiologists, aiding in rapid and accurate COVID-19 detection.