Deep SVDD and Transfer Learning for COVID-19 Diagnosis Using CT Images

Area of Science:

• Medical Imaging and Diagnostics
• Artificial Intelligence in Healthcare
• Infectious Disease Research

Background:

• The COVID-19 pandemic has overwhelmed global health systems, necessitating rapid and accurate diagnostic tools.
• Early diagnosis of COVID-19 is crucial for patient survival and limiting disease transmission.
• Distinguishing COVID-19 from other respiratory illnesses like pneumonia using medical imaging is a significant challenge.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for the automated diagnosis of COVID-19 using chest CT images.
• To differentiate between COVID-19 positive cases, non-COVID-19 pneumonia, and healthy individuals.
• To investigate the efficacy of transfer learning with deep support vector data description (DSVDD) for this diagnostic task.

Main Methods:

• A novel approach combining transfer learning (VGG16 and ResNet50) with one-class deep support vector data description (DSVDD) was proposed.
• Three distinct models were developed, each designed to classify a specific category as normal or anomalous.
• Models were trained using CT image data from multiple sources, curated by an expert radiologist, employing end-to-end fusion and varying data split ratios (70%, 50%, 30%).

Main Results:

• The proposed VGG16-based models achieved F1 scores of 0.8281, 0.9170, and 0.9294 for the different data splits.
• The ResNet50-based models demonstrated strong performance with F1 scores of 0.9109, 0.9188, and 0.9333 across the data splits.
• This represents the first known application of one-class DSVDD combined with transfer learning for diagnosing lung diseases, including COVID-19.

Conclusions:

• The developed transfer learning and DSVDD approach shows high efficacy in distinguishing COVID-19 from other lung conditions on CT scans.
• The method offers a promising tool for early and accurate COVID-19 diagnosis, potentially improving patient management and public health responses.
• This study highlights the potential of advanced AI techniques in addressing critical challenges in medical diagnostics during global health crises.