Quantitative Image-Based Physical Analysis of Lung Microstructure from HRCT for Automated Disease Stratification

Abstract

In this study, quantitative texture analysis of HRCT lung images was performed using MaZda software to classify and compare Tubular Bronchiectasis, COPD, Hydropneumothorax, Pleural Effusion, Hypersensitivity Pneumonitis, and Interstitial Lung Disease (ILD) with normal lung tissue. A total of 127 samples obtained from 41 patients were analyzed to ensure broad representation across diagnostic categories. More than 300 texture features were extracted from selected Regions of Interest (ROIs), and the POE+ACC method was used to identify the ten most discriminative features. These selected features were evaluated using Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Nonlinear Discriminant Analysis (NDA). All three methods achieved 0% misclassification error, demonstrating excellent feature stability and reproducibility. LDA showed strong linear clustering, NDA produced superior nonlinear discrimination using a one-class Artificial Neural Network (ANN), and PCA effectively visualized class variance and consistent grouping. The results confirm that MaZda-based texture analysis offers an accurate, objective, and non-invasive tool capable of distinguishing a wide range of pulmonary diseases using HRCT, supporting radiological decision-making and improving diagnostic precision.