CovC-ReDRNet: A Deep Learning Model for COVID-19 Classification.

Area of Science:

• Computer Science
• Medical Imaging
• Artificial Intelligence

Background:

• The COVID-19 pandemic has caused millions of deaths globally.
• Accurate diagnosis of COVID-19 is crucial but challenging due to clinical similarities with pneumonia.
• Reverse transcription-polymerase chain reaction (RT-PCR) tests are accurate but clinical misdiagnosis persists.

Purpose of the Study:

• To develop a novel deep learning model, CovC-ReDRNet, for distinguishing COVID-19 patients from pneumonia and normal cases.
• To enhance the accuracy and efficiency of COVID-19 diagnosis through an advanced computer-aided diagnostic approach.

Main Methods:

• Developed the CovC-ReDRNet model using ResNet-18 as the backbone for feature representation.
• Employed a feature-based randomized neural network (RNN) framework, integrating deep random vector function link network (dRVFL) as the classifier.
• Validated the model using five-fold cross-validation.

Main Results:

• The CovC-ReDRNet model achieved high performance metrics: 94.94% sensitivity, 97.01% specificity, 97.56% accuracy, 96.81% precision, and 95.84% F1-score.
• Ablation studies confirmed the superiority of ResNet-18 backbone, RNNs over traditional classifiers, and deep RNNs over shallow ones.
• The model surpassed state-of-the-art methods, achieving a maximum accuracy of 97.56% compared to the previous best of 95.57%.

Conclusions:

• The CovC-ReDRNet model demonstrates significant potential as an advanced computer-aided diagnostic tool for COVID-19.
• The model offers high speed and accuracy in classifying and predicting COVID-19, aiding clinical decision-making.
• This deep learning approach addresses the challenge of clinical misdiagnosis between COVID-19 and pneumonia.