Accurate Recognition of Pneumonia and COVID-19 by Geometric Shape Normalization of Lung Region using Automatic Landmark detection and Piecewise Affine Warping

This project presents an automatic classification system for pulmonary diseases like pneumonia and COVID-19 in chest X-ray images based on geometric normalization. The proposed method consists of three main stages. First stage: ResNet-18 convolutional neural networks with coordinate attention mechanism are trained to predict 15 landmarks defining the lung contour, achieving a mean error of 3.61 pixels (median 3.07 pixels) through an ensemble of four models with test-time augmentation. Second stage: the predicted coordinates are used to geometrically normalize each image through Generalized Procrustes Analysis (to obtain a standard lung shape), Delaunay triangulation (to build a deformation mesh), and piecewise affine transformation (warping), eliminating variations in position, scale, and orientation. Third stage: normalized images are classified into three categories (COVID-19, Viral Pneumonia, and Normal) using a ResNet-18 classifier with transfer learning and contrast preprocessing. The system was evaluated on the COVID-19 Radiography Database with 15,153 images, achieving 98.60% accuracy, 97.17% F1-Macro, and 98.09% F1-Weighted on the test set. Five-fold cross-validation on the training and validation sets demonstrated high stability (98.60 +/- 0.26% accuracy), confirming robustness of the approach. Controlled comparison between configurations demonstrates that geometric normalization achieves performance comparable to original images (0.58 percentage points difference), while simple border cropping without normalization produces inferior results.