Improving performance of CNN to predict likelihood of COVID-19 using chest X-ray images with preprocessing algorithms (2006.12229v1)

Published 11 Jun 2020 in eess.IV and cs.LG

Abstract: As the rapid spread of coronavirus disease (COVID-19) worldwide, chest X-ray radiography has also been used to detect COVID-19 infected pneumonia and assess its severity or monitor its prognosis in the hospitals due to its low cost, low radiation dose, and wide accessibility. However, how to more accurately and efficiently detect COVID-19 infected pneumonia and distinguish it from other community-acquired pneumonia remains a challenge. In order to address this challenge, we in this study develop and test a new computer-aided diagnosis (CAD) scheme. It includes several image pre-processing algorithms to remove diaphragms, normalize image contrast-to-noise ratio, and generate three input images, then links to a transfer learning based convolutional neural network (a VGG16 based CNN model) to classify chest X-ray images into three classes of COVID-19 infected pneumonia, other community-acquired pneumonia and normal (non-pneumonia) cases. To this purpose, a publicly available dataset of 8,474 chest X-ray images is used, which includes 415 confirmed COVID-19 infected pneumonia, 5,179 community-acquired pneumonia, and 2,880 non-pneumonia cases. The dataset is divided into two subsets with 90% and 10% of images in each subset to train and test the CNN-based CAD scheme. The testing results achieve 94.0% of overall accuracy in classifying three classes and 98.6% accuracy in detecting Covid-19 infected cases. Thus, the study demonstrates the feasibility of developing a CAD scheme of chest X-ray images and providing radiologists useful decision-making supporting tools in detecting and diagnosis of COVID-19 infected pneumonia.

View on arXiv

Authors (6)

Morteza Heidari (7 papers)
Seyedehnafiseh Mirniaharikandehei (3 papers)
Abolfazl Zargari Khuzani (6 papers)
Gopichandh Danala (3 papers)
Yuchen Qiu (6 papers)
Bin Zheng (32 papers)

Citations (333)

View on Semantic Scholar

Summary

Enhancing CNN Performance for COVID-19 Diagnosis using Chest X-ray Images through Preprocessing Algorithms

The paper presented explores the application of convolutional neural networks (CNN), particularly the VGG16 model, in developing a computer-aided diagnosis (CAD) system for the classification of chest X-ray images. The research aims to address the challenges in accurately identifying COVID-19 infected pneumonia, alongside other community-acquired pneumonias and normal (non-pneumonia) cases, leveraging pre-processing techniques for enhanced model performance.

Methodological Approach

The research leverages a VGG16-based CNN model, employing transfer learning to mitigate the limitations associated with training from scratch on a relatively small, unbalanced dataset. The VGG16 architecture, initially trained on the extensive ImageNet database, provides a robust foundation for fine-tuning with chest X-ray images for this specific task. The dataset comprises 8,474 images, divided into 415 COVID-19 cases, 5,179 instances of other pneumonias, and 2,880 normal cases, with a 90/10 split used for training/testing respectively.

Image preprocessing before CNN input is a major innovation in this paper. The preprocessing includes segmenting and removing the diaphragm region, noise reduction through bilateral filtering, and normalization of contrast using histogram equalization. These steps generate three input images per original chest X-ray, corresponding to the VGG16 model's three RGB input channels, despite the images being grayscale, which effectively enhances model capability by incorporating varied texture information.

Results and Performance

The final CNN model exhibits substantial accuracy improvements, achieving an overall accuracy of 93.9% in distinguishing among the three classes and a remarkable 98.6% accuracy in specifically detecting COVID-19 pneumonia. These are supported by a Cohen's kappa score of 0.88, indicating high reliability and robustness of the classification results. These figures highlight the efficacy of the preprocessing techniques and the VGG16-based transfer learning approach. Additionally, a comparison of results with models omitting preprocessing steps demonstrates significant performance drops, underscoring preprocessing's critical role.

Implications and Future Directions

The paper's findings have practical implications, indicating that such a CAD system can substantially aid radiologists by streamlining the diagnostic process of differentiating COVID-19 from other forms of pneumonia using accessible imaging techniques like chest X-rays. The research demonstrates the potential for similar methodologies to be applied broadly in medical image classification tasks beyond respiratory disease diagnostics.

Future research could expand upon these findings by exploring additional preprocessing methods, such as more sophisticated segmentation algorithms, or integrating other deep learning architectures. The paper's reliance on a specific dataset invites further validation and testing across diverse and larger datasets to solidify and generalize these promising results. The exploration of alternative and complementary machine learning approaches in tandem with CNNs also warrants exploration for enhanced robustness and accuracy across varying clinical scenarios.

Overall, this research offers valuable insights into developing optimally-tuned CNN models for complex medical imaging tasks, contributing to the broader field of AI-driven diagnostics in healthcare.

PDF Markdown

Related Papers

Find Related Papers