Deep Learning Morphological Analysis of Chest Radiographs as Preliminary Predictor for Hospitalization in Patients with Chronic Cardiac Disease

Abstract

Objectives: This study uses deep learning to analyze morphological features of chest radiographs to predict the need for hospitalization in patients with cardiac problems. Methods: A retrospective case-control approach was designed. A public dataset of chest x rays was used and validated. The multimodal AI model was designed to extract x-ray features at different CNN levels to classify patients into 'admission' and 'No admission' groups. Results: Several important morphological features from patients attributes and chest radiographs were obtained to predict the need for hospitalization in patients with chronic cardiac disease. These preliminary predictors, belonging to different CNN layers: L3 (layer 3) Texton Density 61, L4 Reflection Symmetry 40, Relative Size Context may be linked to structural changes and severity. Conclusion: There was a preliminary association between multimodal DL features and personalized hospitalization decisions in patients with chronic cardiac disease.

Ähnliche Arbeiten