Yapay Zeka Yöntemi ile Bölütlenmiş Karmaşık Damar Yapılarının Üç Boyutlu Biyoyazıcı ile Üretimi

Abstract

According to the World Health Organization, heart disease is one of the leading causes of death worldwide. Computed Tomography Angiography (CTA), which is an interventional medical imaging method and very detailed images can be obtained, is used in the diagnosis and treatment planning of heart and coronary diseases. Computer vision and computerized diagnostic support systems are needed to examine coronary arteries through CTA images. Withing the scope of this thesis, the segmentation of coronary arteries with artificial intelligence and the bioprintability of the three dimensional (3D) model obtained after segmentation were examined. In this context, the training was carried out with 3D U-Net in nnU-Net, which is a deep learning architecture framework, using CLS12 CTA images, in which publicly accessible coronary arteries are classified. Three dimensional segmentation of the complex coronary arteries was performed fully automatically with U-Net, MultiRes U-Net, nnU-Net, and nnU-net with transfer learning, and compared with the others, nnU-net training with transfer learning was better with a Dice similarity coefficient (DSC) of 0.86. provided a successful segmentation. As a result of segmentation, 3D coronary artery model was successfully obtained, saved in '.stl' format and made available with a 3D bioprinter.

In the second part of the thesis, an available 3D printer working with the melt deposition method was converted into a 3D bioprinter at low cost using open source hardware and software tools for bioprinting the 3D coronary artery model created by artificial intelligence. The developed coronary artery model was bioprinted into the Pluronic F-127 support bath by a 3D bioprinter working with a microextrusion principle using the alginate bioink. In the continuation of the study, 3D bioprinted vascular structures were characterized by macroscopic imaging, micro-computed tomography ($\mu$-CT) and flow experiments. In these analyzes, it was observed that the bioprinted artery structure could be produced as hollow and the solution flow sent into it was regular and impermeable. In macroscopic and CT imaging analyzes, 3D models created over ground truth and artificial intelligence results were compared. Analyzes made on the macroscopic images of the bioprinted structures at a scale of 200\%, the vessel diameters were measured as 4.3 mm and 6 mm in different regions for the ground truth 3D vessel structure, and 4 mm and 5.8 mm for the artificial intelligence output 3D vessel structure. In the analyzes made on $\mu$-CT imaging of freeze-dried samples, the outer vessel diameters are 4 mm and 5.3 mm in different regions and the inner lumen diameter is 2 mm and 1.5 mm in different regions for the ground truth 3D vessel structure, the outer vessel diameters are 3.5 mm and 5 mm in different regions and the inner lumen diameter is 1.8 mm and 1.5 mm in different regions for the artificial intelligence 3D vessel structure.

As a result, within the scope of this thesis, personalized 3D coronary artery segmentation with an artificial intelligence method has been successfully performed independently of the user and the manufacturability of complex structures with a 3D bioprinter has been demonstrated. It has been shown that 3D coronary arteries segmented via artificial intelligence can be successfully used in vascular tissue engineering studies with microextrusion-based bioprinters. Moreover, when three-dimensional bioprinting processes are supported by artificial intelligence, it is thought that it can guide the development of other tissues/organs in complex architecture.