An automatic computer-aided detection scheme for pneumoconiosis on digital chest radiographs

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Occupational Health

Background:

• Pneumoconiosis poses a significant public health challenge, necessitating efficient and accurate diagnostic tools.
• Current diagnostic methods for pneumoconiosis can be labor-intensive and subjective, highlighting the need for automated solutions.

Purpose of the Study:

• To develop and evaluate an automated computer-aided detection (CAD) scheme for pneumoconiosis detection using digital chest radiographs.
• To improve the accuracy and efficiency of pneumoconiosis screening through an automated system.

Main Methods:

• Lung fields were segmented using active shape models and subdivided into six regions based on Chinese diagnostic criteria.
• A multi-scale difference filter bank enhanced small opacity details, and texture features were extracted from original and processed images.
• Support vector machine classifiers were employed for classification, with final decisions based on chest-level reporting and regional probabilities.

Main Results:

• The automated CAD scheme demonstrated high classification performance on a dataset of 300 normal and 125 pneumoconiosis cases.
• Selected feature vectors resulted in high classification accuracy, outperforming previous methods.
• The system achieved higher accuracy and more convenient interaction compared to existing approaches.

Conclusions:

• The developed automated CAD scheme is effective for detecting pneumoconiosis on digital chest radiographs.
• This system offers a valuable tool for the mass screening of pneumoconiosis in clinical settings.
• The scheme's high accuracy and user-friendliness contribute to improved early detection and management of pneumoconiosis.