Lung pErfusioN Analysis with big data techniques

• Applicant: Dr. Hinrich Winther
• Subject Area: Nuclear Medicine, Radiotherapy, Radiobiology
• Term: from 2016 to 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 314828988

The World Health Organization lists chronic obstructive pulmonary disease (COPD) as the third most common cause of death. The Centers for Disease Control and Prevention of the United States of America expects a national cost increase from an overall $59.3 billion in 2010 to $90.6 billion by 2020. Thus, it is of utmost importance to identify and monitor COPD adequately in order to properly treat patients to inhibit the progression of the disease and to reduce treatment costs. Currently, the clinical lung function test is used to diagnose COPD. However, this technique is fairly insensitive to detect regional, early change of COPD and does not quantify emphysema.Several multicenter studies, such as COPD Gene, ECLIPSE, SPIROMICS and MESA COPD are searching for predictive biomarkers for COPD patient outcome. Impairment and alteration of perfusion were identified as important features. Non-invasive MRI techniques can reliably quantify regional lung perfusion and provide important information. However, implementing quantitative measurement techniques in the clinical setting requires human interaction, running the risk of high inter-observer variability and binding valuable human resources.The MESA COPD Study provides the data set for this proposal. It enlisted 4.617 patients, which were extensively characterized. CT and MR images were obtained, allowing quantitative measurements of parenchymal microvascular pulmonary blood flow (PBF). To ensure optimal PBF measurement in the microvasculature of lung parenchyma conducting vessels, bullae and emphysema lung regions have to be excluded. While the conducting vessels can be seen on the MRI data there is currently no adequate measure of emphysema without CT data. This information needs to be mapped on to the perfusion MRI data set.The process of evaluating a large image data sets per patient requires lengthy and error prone manual interaction. It is not feasible to examine this amount of image data. Therefore, we propose the development of an algorithm, based on big data techniques, to automatically perform these measurements.We will evaluate the applicability of automatic segmentation techniques in order to measure parenchymal microvascular PBF more precisely. Additionally, we will establish a parameter set for the "vulnerable zones" of the lung parenchyma, identifying highly endangered regions. To achieve this task big data techniques, such as machine learning and image registration, will be applied. Preliminary work shows initial promising results in automatic image segmentation by machine learning as well as image registration.The impact of this work is a milestone for new image-derived parameters using big data techniques to detect early change of COPD lung disease and define novel predictive biomarkers for patient outcome, far beyond the current clinical lung function test.

• DFG Programme: Research Grants