COVID-19 diagnosis using a decision tree algorithm

Coronavirus disease 2019 (COVID-19) is a respiratory infection for which chest X-ray (CXR) imaging can provide an accessible screening modality. This study investigates an interpretable computer-aided screening pipeline for distinguishing COVID-19-positive from normal CXR images. Methods: The pipeline combines Otsu-based lung segmentation, morphological refinement, seven Hu invariant shape moments, training-fold z-score normalization, and a class-weighted Decision Tree. A cohort of 2,600 CXR images (1,000 COVID-19-positive and 1,600 normal) was evaluated using stratified five-fold cross-validation. The retained experimental record reports 98.6% accuracy, 97.8% precision, 98.6% recall (sensitivity), and 98.2% F1-score. The pooled confusion matrix shows TP=986, FN=14, FP=22, and TN=1578, corresponding to 98.6% accuracy and 97.8% precision. ROC/AUC analysis yielded AUC = 0.987 ± 0.004 under the image-level evaluation protocol. Conclusion: The findings indicate that a low-dimensional, shape-based descriptor combined with an interpretable Decision Tree can discriminate COVID-19-positive from normal CXR images under the present internal image-level protocol. Because the cohort was not partitioned at the patient or source level and no independent external cohort was retained in the study record, the results are considered preliminary proof-of-concept evidence rather than clinically validated diagnostic performance.