Radiographic severity and treatment optimisation in Mycobacterium avium complex pulmonary disease

Area of Science:

• Pulmonology
• Medical Imaging
• Artificial Intelligence

Background:

• Mycobacterium avium complex pulmonary disease (MAC-PD) requires accurate severity assessment for effective treatment.
• Chest radiography (CXR) is a common imaging modality for diagnosing pulmonary diseases.
• Deep neural networks (DNNs) offer potential for quantitative image analysis in medicine.

Purpose of the Study:

• To develop and validate a DNN-based scoring system to quantify MAC-PD severity using CXRs.
• To stratify patients with MAC-PD into mild and extensive disease categories based on CXR scores.
• To evaluate the impact of different treatment strategies on culture conversion rates in relation to disease severity.

Main Methods:

• A DNN was developed to generate a CXR score for MAC-PD severity.
• Chest radiographs of 541 patients with MAC-PD were analyzed, classifying them into mild (score ≤ 4.0) or extensive (score > 4.0) disease.
• Cox proportional hazards models assessed the effect of macrolide/ethambutol-based, alternative intensified, and intensified regimens on culture conversion.

Main Results:

• The study included 541 patients: 299 with mild and 242 with extensive MAC-PD.
• Overall culture conversion was achieved in 83.2% of patients, with no significant difference between mild (81.6%) and extensive (85.1%) groups.
• In extensive MAC-PD, intensified treatment regimens (including IV amikacin) significantly improved culture conversion (aHR 1.90, 95% CI 1.12-3.22), while this was not observed in mild disease.

Conclusions:

• A DNN-based CXR scoring system effectively stratifies MAC-PD disease extent.
• This stratification can guide individualized treatment decisions for MAC-PD patients.
• Intensified treatment regimens are crucial for improving outcomes in patients with extensive MAC-PD.