Author: John P. Aris, Wesley E. Bolch, Robert Joseph Dawson, Bonnie N. C. President, Yitian Wang 👨🔬
Affiliation: Johns Hopkins University, University of Florida 🌍
Purpose: Generation of a mesh-based microscale lung model is essential for accurate dosimetry analysis. Lungs exchange air with the environment and may be exposed to alpha-particle-emitting radionuclides present in aerosols, which can cause significant damage to lung tissue. To evaluate the damage in lung tissue, developing a microscale lung model is a crucial yet challenging task in computational dosimetry. With the recent advancements in generative AI, it is now possible to develop microscale lung models more efficiently and accurately. Once the microscale lung model is assembled, radionuclide S-values are calculated and applied for dosimetry application.
Methods: A set of 25 human elastic-stained (Iodine) histology slides were inspected under microscopic viewing and assembled consecutively. Interpolated images were generated between each slice using the generative AI tool, FILM, which is running in the HiPerGator, the University of Florida's supercomputer system. Segment Anything Model (SAM), another generative AI tool, was employed to segment large, independent structures, including blood vessels and major airways, within the histology images. The segmented components were then imported into 3D Slicer to generate a 3D microscale mesh lung model, with manual corrections applied to address inaccuracies. Finally, the mesh models were imported into 3D rendering software, to reduce the model size and correct errors in the 3D structure. Other structures, like the alveolar region, will be generated using in-house python or C++ code. The mesh models can be used to simulate monoenergetic alpha particles in lung tissue and calculate the S values for different radionuclides.
Results: Seven different microscale lung tissue models were successfully generated, encompassing multiple lung regions. Computation of S-values for several radionuclides was completed, including Cs-137.
Conclusion: A set of polygon mesh microscale lung models was constructed with AI assistance, and the corresponding analyses can be applied to a variety of dosimetry applications, including radiopharmaceutical studies.