COVID-ResNet: A Deep Learning Framework for Screening of COVID19 from Radiographs (2003.14395v1)

Abstract: In the last few months, the novel COVID19 pandemic has spread all over the world. Due to its easy transmission, developing techniques to accurately and easily identify the presence of COVID19 and distinguish it from other forms of flu and pneumonia is crucial. Recent research has shown that the chest Xrays of patients suffering from COVID19 depicts certain abnormalities in the radiography. However, those approaches are closed source and not made available to the research community for re-producibility and gaining deeper insight. The goal of this work is to build open source and open access datasets and present an accurate Convolutional Neural Network framework for differentiating COVID19 cases from other pneumonia cases. Our work utilizes state of the art training techniques including progressive resizing, cyclical learning rate finding and discriminative learning rates to training fast and accurate residual neural networks. Using these techniques, we showed the state of the art results on the open-access COVID-19 dataset. This work presents a 3-step technique to fine-tune a pre-trained ResNet-50 architecture to improve model performance and reduce training time. We call it COVIDResNet. This is achieved through progressively re-sizing of input images to 128x128x3, 224x224x3, and 229x229x3 pixels and fine-tuning the network at each stage. This approach along with the automatic learning rate selection enabled us to achieve the state of the art accuracy of 96.23% (on all the classes) on the COVIDx dataset with only 41 epochs. This work presented a computationally efficient and highly accurate model for multi-class classification of three different infection types from along with Normal individuals. This model can help in the early screening of COVID19 cases and help reduce the burden on healthcare systems.

• The paper introduces a CNN framework named COVID-ResNet that extends ResNet-50 with progressive resizing, achieving 96.23% accuracy on the COVIDx dataset.
• The methodology employs cyclical and discriminative learning rates over three training stages, reducing training epochs to 41 while optimizing performance.
• The results demonstrate perfect sensitivity and PPV for the COVID-19 class, highlighting its potential to reduce diagnostic reliance on PCR tests.

COVID-ResNet: Screening COVID-19 via Deep Learning

The paper "COVID-ResNet: A Deep Learning Framework for Screening of COVID-19 from Radiographs" introduces a novel approach to accurately identify COVID-19 infections using chest X-rays. The authors, Muhammad Farooq and Abdul Hafeez, design a Convolutional Neural Network (CNN) framework termed COVID-ResNet to enhance classification accuracy between COVID-19 and other pneumonia types.

Methodology

The core of the paper lies in the utilization of the COVIDx dataset, comprising 5941 chest X-ray images across four classes: Normal, Bacterial pneumonia, Viral pneumonia, and COVID-19. The challenge is compounded by the imbalanced sample distribution, with COVID-19 cases being underrepresented. The COVID-ResNet framework extends the pretrained ResNet-50 architecture by progressively resizing input image dimensions to improve learning efficiency. This strategy, coupled with advanced techniques like cyclical learning rates and discriminative learning rates, aids in achieving superior performance.

Progressive resizing allows the model to leverage information at multiple resolutions, enhancing generalization. The training process unfolds across three stages, each employing different image sizes, ultimately resulting in robust feature learning. An adaptive approach in learning rate selection further optimizes the training duration, culminating in a mere 41 epochs.

Results

COVID-ResNet demonstrated a notable improvement over previous models like COVID-Net, achieving an accuracy of 96.23% on the COVIDx dataset. This is a significant step forward, considering the prior report of 83.5% accuracy by COVID-Net. Furthermore, the model efficiently balances performance with computational complexity, as evidenced by a 4.5-fold reduction in parameters (25.6M compared to 116.6M).

From a class-wise perspective, COVID-ResNet achieves perfect sensitivity and positive predictive value (PPV) for the COVID-19 class, underscoring its diagnostic potential. However, it also performs well across other categories, as reflected in its impressive F1-scores and minimal misclassification, underscoring robust feature extraction.

Implications and Future Directions

The paper’s findings imply that COVID-ResNet can potentially reduce diagnostic over-reliance on PCR tests, offering a fast and reliable preliminary screening tool. This can alleviate pressure on healthcare infrastructures by swiftly identifying true COVID-19 cases for further testing. The methodological advancements in progressive resizing and learning rate strategies exemplify scalable solutions that calibrate model training with performance effectively.

While COVID-ResNet exhibits promising potential, future work must consider scaling up with larger and more diverse datasets to ensure clinical viability. Additionally, incorporating explainability measures could demystify the decision pathways, fostering greater trust and adoption in medical contexts.

Overall, this work underscores the viability of open-access, AI-driven diagnostic tools, pushing the frontiers of rapid and accurate disease detection in pandemic scenarios.