Artificial Intelligence Assisted Detection of Respiratory Infectious Diseases Signs From Computed Tomography Images

Faezeh  Shalbafzadeh; Fatemeh  Taherpour-Dizaji; Mohammad Reza  Fouladi; Sharareh  Baradaran; Matin  Ghadiri; Hossein  Ghadiri

doi:10.18502/fbt.v13i1.20336

Faezeh Shalbafzadeh Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran
Fatemeh Taherpour-Dizaji Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran
Mohammad Reza Fouladi Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran
Sharareh Baradaran Department of Radiology, West Nikan Hospital, Tehran, Iran
Matin Ghadiri School of Electrical and Computer Engineering,University of Tehran, Tehran, Iran
Hossein Ghadiri Department of Medical Physics and Biomedical Engineering, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran

DOI: https://doi.org/10.18502/fbt.v13i1.20336

Keywords: Respiratory Infectious Diseases; Artificial Intelligence; Random Forest Algorithm; Weka; Chest Computed Tomography Images.

Abstract

Purpose: Respiratory infectious diseases often manifest as Ground-Glass Opacity (GGO) or consolidation signs in the lungs. Artificial Intelligence (AI)-assisted systems utilizing data mining algorithms such as Waikato Environment for Knowledge Analysis (Weka) can be used for the detection and segmentation of these signs. In this study, we propose using Weka as a comprehensive data mining and machine learning tool to develop the most accurate models for detecting lung signs in chest CT images of patients with respiratory infectious diseases.

Materials and Methods: First, we manually selected specific signs from chest Computed Tomography (CT) images from 600 cases using the Graphical User Interface (GUI) Weka plugin. We then trained the random forest algorithm based on different features and presented the best combined model obtained for the automatic detection of the aforementioned signs. Lastly, the model's performance was evaluated with different metrics.

Results: Our findings indicate that the hybrid texture description features, including “Structure”, “Entropy”, “Maximum”, “Anisotropic”, and “Laplacian” available in Weka, demonstrated the lowest Out-of-Bag (OOB) error rate, highest Area Under the ROC Curve (AUC) value of 0.992, and accuracy of 98.1%.

Conclusion: By leveraging the combination of Weka features, we have successfully developed models for the detection and segmentation of lung signs associated with infectious diseases, from chest CT images. These findings contribute to the field of medical image analysis and hold promise for improving the diagnosis and treatment outcomes of patients with respiratory infectious disorders.