ARTIFICIAL INTELLIGENCE BASED ON DEEPLUNGS MODELS AND CHEST X-RAY IMAGING

Abstract

Introduction: Artificial intelligence (AI) is one of the new directions, which received a new impetus in development after the COVID-19 pandemic.

 Materials and method: Cross-sectional analysis of an observational study conducted in the radiology department of a multidisciplinary clinic in Zhambyl region in 2023. Chest X-rays were performed on 1133 patients aged 18 to 77 years, and the radiographs were read twice by two independent radiologists and DeepLungs Models. The cascade of models identifies 14 symptoms: Inflammatory infiltration (ROC AUC 73%), COVID19; Pneumonia (ROC AUC 79%), pneumonia; Focal lung opacification (ROC AUC 90%); Pulmonary nodules (ROC AUC 84%); Effusion (ROC AUC 90%); Pleural thickening (ROC AUC 85%); Atelectasis (ROC AUC 85%); Cardiomegaly (ROC AUC 91%); Pneumothorax (ROC AUC 91%), pneumothorax; Pulmonary consolidation (ROC AUC 84%); Edema (ROC AUC 93%); Emphysema (ROC AUC 92%); Fibrosis (ROC AUC 86%); Hernia (ROC AUC 94%).

 Results: Males 175 (45.6%) Females 209 (54.4%), P value 0.086. Patients with clinical manifestations of respiratory diseases in 169 (44.0%). X-ray semiotics of pneumonia in 21 (5.5%). The DeepLungs Models “Pneumonia” system had ≥90% confidence in identifying of pneumonia in 43 (11.2%) patients, P value 0.465. The specificity of the DeepLungs Models Pneumonia method is 77.5% CI95% [70.9%; 83.2%], sensitivity - 14.3% [9.1%; 21.0%], positive predictive value - 32.8% [23.3%; 44.0%], the predictive value of a negative result is 54.1% [51.5%; 56.5%].

 Conclusions: Despite current advances, further work is needed to improve the diagnostic accuracy of DeepLungs Models in detecting respiratory diseases.