Contour-based lung shape analysis in order to tuberculosis detection: modeling and feature description

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Statistical Shape Analysis

Background:

• Pulmonary diseases like tuberculosis (TB) require reliable diagnostic methods.
• Statistical shape analysis offers a promising alternative for disease diagnosis.
• Existing lung shape analyses primarily focus on content-based methods.

Purpose of the Study:

• To develop and investigate a 2D contour-based lung shape analysis for diagnosing pulmonary tuberculosis (PTB).
• To model normal and abnormal lung shapes using Fourier descriptors and principal component analysis.
• To evaluate the effectiveness of contour-based features for TB detection.

Main Methods:

• Utilized Fourier descriptors (FDs) to represent lung shape models from chest X-ray (CXR) images.
• Applied principal component analysis (PCA) to analyze inter-patient lung shape variations.
• Trained a classifier using contour-based feature vectors on the Montgomery dataset for TB detection.

Main Results:

• The proposed approach explained over 95% of shape variations using 6-7 principal component modes for right and left lungs.
• Achieved 82.03% accuracy and 88.75% AUC for PTB detection in a dataset of 138 lung cases.
• Demonstrated robustness and value for 2D automatic segmentation, classification, and atlas registration.

Conclusions:

• Contour-based lung shape analysis using FDs and PCA is a promising method for PTB diagnosis.
• The approach offers a valuable supplement to existing diagnostic tools for pulmonary diseases.
• The method is adaptable for various pulmonary conditions and image analysis tasks.