Multivariate analysis of radiological signs for the differential diagnosis of pulmonary infections

This study addressed the challenging task of etiological diagnosis in pulmonary infections, which requires the integration of epidemiological knowledge, clinical history, and radiological semiology. The aim was to quantify the diagnostic value of radiological findings described in the literature for differentiating pulmonary infections. Radiological features of 39 pulmonary infections were reviewed, including 19 bacterial, 13 fungal, and 7 viral etiologies; 56 publications were analyzed. Overall, six imaging patterns and 27 radiological signs were identified. Multivariate analysis was performed using Factor Analysis of Mixed Data (FAMD) to define a diagnostic space, followed by cluster analysis of diagnoses and radiological signs, and least absolute shrinkage and selection operator (LASSO)-regularized logistic regression models. Within the FAMD diagnostic space, Dimension 1 separated chronic and indolent infections-mainly mycobacterial and fungal infections-from acute bacterial and viral pneumonias, whereas Dimension 2 distinguished bacterial from viral etiologies. Cluster analysis identified five major groups: (1) non-tuberculous mycobacterial infections, (2) fungal and fungoid bacterial infections, (3) viral infections and atypical pneumonias, (4) typical bacterial pneumonias, and (5) primary tuberculosis and other specific infections. Comparison between acute bacterial and chronic/fungal infections showed moderate separation. When acute bacterial and viral infections were compared, models based on radiological signs showed limited-to-moderate discrimination; however, imaging patterns alone substantially improved diagnostic performance. Interstitial-alveolar and reticulonodular patterns were strongly associated with viral etiology. This finding highlights the diagnostic value of global pattern recognition, consistent with the radiological principle that diffuse, low-contrast abnormalities are better appreciated when the chest radiograph is viewed from a distance or through a minifying lens. These findings provide quantitative evidence supporting the classical semeiological approach to the differential diagnosis of pulmonary infections, reinforcing the complementary role of individual radiological signs and global imaging patterns in chest imaging.