.png){kind=icon}
.png){kind=icon}
Return
Construction of human bronchial trees from computed tomography scans through machine learning and stochastic growth algorithms
.png){kind=icon}
OA
AI
G
A
T
F
C
A
DOI：10.1093/jcde/qwag050.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Digital twins of human lungs are of great value for healthcare, pharmaceutical and bio-engineering communities. They can be used to study ventilation and gas mixing problems informing and improving the design of assisted ventilation devices. They can also provide an estimate of the occupational exposure to toxic or pathogenic aerosols. Furthermore, they can be employed to optimize therapeutic aerosols and their delivery through inhalers or nebulizers. Fundamental for all these applications is the possibility to construct accurate and physiologically realistic models of bronchial trees. We developed a machine learning-based method to extract the 3D structure of the upper part of the bronchial tree to the 4th-5th generation, and the 3D lung lobe volumes from a computed tomography scan. The unresolved part of the bronchial tree down to the terminal alveolar sacs, 6th to 23rd generations, is then generated using a constrained stochastic growth algorithm. Our method is discussed in detail and its performance compared to similar algorithms available in the literature. The quality and physiological representativity of the generated trees are analyzed and a comparison made with the available morphometric data. To conclude, the current limitations and the further developments towards a realistic virtual lung model are illustrated.
Keywords:
bronchial tree
digital twin
machine learning
stochastic growth
computed tomography
Journal
IF:
6.1
Papers:
392
Citations:
3.2K
Organization
Cited Papers