Radiographic severity and treatment optimisation in

Background: We developed a chest radiography (CXR) scoring system using a deep neural network to quantify the severity of Methods: We analysed chest radiographs obtained at treatment initiation in patients who began therapy for MAC-PD at Seoul National University Hospital between 1 January 2011, and 31 August 2024. Patients were classified into mild or extensive disease groups based on a CXR score cut-off of 4.0. The impact of macrolide/ethambutol-based initial treatment regimens, alternative intensified regimens (including inhaled amikacin or clofazimine), and intensified regimens (including intravenous amikacin) on culture conversion was assessed using Cox proportional hazards models. Results: The study comprised 541 patients (299 and 242 with mild and extensive disease types, respectively). Alternative intensified regimens were used in 103 patients (19.0%), and intensified regimens were used in 25 patients (4.6%). Culture conversion was achieved in 450 patients (83.2%), with no significant differences between the mild (81.6%) and extensive (85.1%) groups (p=0.277). In mild disease, neither alternative intensified regimens (adjusted hazard ratio (aHR) 0.84, 95% confidence interval (CI) 0.60-1.19) nor intensified regimens (aHR 0.99, 95% CI 0.53-1.89) were associated with improved culture conversion. In contrast, in extensive disease, use of intensified regimens (aHR 1.90, 95% CI 1.12-3.22) was associated with a higher likelihood of culture conversion. Conclusion: CXR score-based disease stratification may guide individualised treatment strategy in patients with MAC-PD.