Radiomics and Deep Learning for Analysis of the Pulmonary Vasculature on MR Imaging in the German National Cohort (NAKO)

• Applicant: Dr. Ricarda von Krüchten
• Subject Area: Radiology
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 519189125

Chronic obstructive pulmonary disease (COPD) is one of the leading causes of death worldwide. Smoking is the most common risk factor. The development and progression of the disease is a gradual, irreversible, and dangerous process. As the disease progresses, the small vascular structures in the lungs responsible for gas exchange are increasingly damaged. Due to this remodeling of the vessels, the blood in the pulmonary arteries can be obstructed in its flow, and the pressure in this vessel subsequently increases. This so-called pulmonary hypertension occurs in most COPD patients and is associated with a poor prognosis. Consequently, there is a great interest in methods for early detection and possibilities to stratify patients according to their risk to initiate individual treatments as early as possible. In clinical routine, lung function tests are used to monitor COPD patients. However, these are not predictive of early detection and risk stratification. Similarly, imaging studies cannot adequately assess vascular structures to estimate the risk. Artificial intelligence offers an opportunity to use non-invasive biomarkers to diagnose, predict prognosis and monitor treatment success. The assessment of vascular structures and lung volumes using radiomics requires complex segmentation. To become relevant for clinical routine, automated segmentation of the vessels must therefore be possible. The whole-body MRI data in the German National Cohort (NAKO) constitutes a unique, promising opportunity for developing auto-segmentation of lung volumes and pulmonary vessels. In this study, an algorithm will be developed and validated that reliably detects lung volumes and vascular structures automatically. In a second step, radiomic signatures will be correlated with lung volumes. As our preliminary work has shown that lung volumes correlate with pulmonary function tests, this study offers a great chance to identify features within the vascular system to draw conclusions regarding lung function. In summary, this study offers a unique and promising opportunity to guide the treatment of critically ill patients in an individualized and targeted manner to improve survival and quality of life substantially.

• DFG Programme: Research Grants
• Co-Investigators: Dr. Tobias Norajitra; Professor Dr. Christopher Schlett