RADIC:A tool for diagnosing COVID-19 from chest CT and X-ray scans using deep learning and quad-radiomics

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Deep learning (DL) models show promise for rapid COVID-19 diagnosis using CT and X-ray scans.
• Current DL models primarily use spatial image information, potentially limiting diagnostic accuracy.
• Radiomics-generated images and combined feature analysis may enhance DL model performance for COVID-19 detection.

Purpose of the Study:

• To develop and validate RADIC, an automated tool for COVID-19 detection using DL models trained on radiomics-generated images.
• To investigate the efficacy of combining radiomics features with time-frequency representations for improved diagnostic accuracy.
• To compare the performance of RADIC against existing studies for COVID-19 detection.

Main Methods:

• Four radiomics approaches were used to analyze original CT and X-ray images.
• Three DL models were trained on distinct sets of radiomics, X-ray, and CT images.
• Deep features were extracted, dimension-reduced via Fast Walsh Hadamard Transform, and combined using discrete cosine transform for classification.

Main Results:

• RADIC achieved high diagnostic accuracy: 99.4% on the CT dataset and 99% on the X-ray dataset.
• The study demonstrated that DL models trained with radiomics-generated images outperform those trained with original scans.
• Incorporating deep features from multiple radiomics approaches significantly improved diagnostic accuracy.

Conclusions:

• RADIC offers a highly accurate and automated solution for COVID-19 detection.
• Training DL models with radiomics-generated images is more effective than using original CT and X-ray images.
• The integration of multi-radiomics deep features enhances the diagnostic performance for COVID-19 detection.