Computer Aided Detection of SARS Based on Radiographs Data Mining

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Data Mining

Background:

• Severe acute respiratory syndrome (SARS) poses a significant public health challenge.
• Accurate and timely detection of SARS is crucial for effective patient management and disease control.
• Traditional diagnostic methods for SARS can be time-consuming and may require specialized expertise.

Purpose of the Study:

• To develop and evaluate an automated computer-aided detection (CAD) system for identifying SARS using image mining techniques.
• To explore the efficacy of different classification algorithms in detecting SARS from chest X-ray images.
• To establish a prototype CAD system for potential clinical application in SARS detection.

Main Methods:

• Utilized digitalized posterior-anterior (PA) X-ray images from a Picture Archiving and Communication System (PACS).
• Employed image texture classification methods after initial association rule mining proved inconclusive.
• Trained and tested classification models including C4.5, Neural Network (NN), and Classification and Regression Trees (CART) on a dataset of SARS and pneumonia X-ray images.

Main Results:

• Classification based on image textures was performed using C4.5, NN, and CART algorithms.
• The CART algorithm demonstrated the highest performance, achieving 70.94% accuracy in detecting SARS cases.
• Receiver Operating Characteristic (ROC) charts and confusion matrices were generated and analyzed for all three methods.

Conclusions:

• Image mining techniques, particularly texture-based classification with CART, show promise for automated SARS detection.
• The developed CAD system prototype offers a potential tool for assisting clinicians in identifying SARS from X-ray images.
• Further research and validation are warranted to enhance the system's accuracy and clinical utility.