Undergraduate Senior Design Project
Application of Personalized Airway Trees in Multi Scale Lung Models to Probe Structure-Function Relations in Asthmatics
Summary:
As an undergraduate student at the Respiratory Physiology and Systems Identification Laboratory at BU, I created a lung model for analyzing mechanics on a patient specific basis using hyperpolarized helium MRI images. The results of this project lead to a publication in the Medical Image Computing and Computer-Assisted Intervention (MICCAI) conference in Copenhagen, Denmark.
Download Full Report (2.42MB)
Download 21st Annual Senior Project Conference Presentation (1.66MB)
Download Undergraduate Research Symposium Poster (0.33MB)
Download MICCAI Publication (0.22MB)
Download MICCAI Poster (0.26MB)
Abstract:
The human lungs are a complex system of bifurcating airways. Diseases affecting the airways result in the loss of lung function. Previous computational modeling work by Tgavalekos et al invoked a concept called Image-Function Modeling (IFM). The IFM approach uses a generic three-dimensional lung model and ventilation images (MRI or PET) to predict which airways in the model contribute to the degradation of lung function. The IFM method would be more robust if the 3D lung model was personalized for each subject for which it predicts airway closures. The overall goal of this project is to design and apply software that generates patient specific airway tree models. To obtain patient specific models, Hyperpolarized Helium MRI Imaging (Hyp. 3He MRI) of the lungs was used (about 13 coronal images that are 256 by 256 pixels) to extract data boundaries of a 3D lung for an individual subject. However, divisions between individual lung lobes (fissures) are not visible. Image registration of cryosection images of a frozen cadaver (Visible Human Data (VHD), US National Library of Medicine) was performed to map the fissure locations. The lung fissure locations are visible on the high resolution cryosection images (2048 by 1216 pixels). Collaborative efforts with Betke et al in the Computer Science department at Boston University have implemented the registration procedures necessary for mapping fissure locations from the VHD images into the MRI images. The algorithm generates an airway tree into the personalized lung space. With these trees, various constriction patters (means and standard deviations ranging from 0 to 60 percent with 20 percent increments) were imposed, and we predicted lung resistance and elastance versus frequency. Prediction of resistance and elastance from our personalized airway trees show that trees are consistent with but clearly distinct from the original Tawhai model. The distribution of airways in the model was skewed. Sensitivity tests revealed that the skewed result cannot be avoided when airway termination parameters are varied. These findings suggest that personalized models may prove critical improvements to pattern matching simulations with real oscillatory mechanics and imaging. Future studies need to establish if predicting ventilation and mechanics for an individual will be distinct from the personalized approach.