Date of Award
12-2023
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mechanical and Civil Engineering
First Advisor
Linxia Gu
Second Advisor
Darshan G. Pahinkar
Third Advisor
Ryan T. White
Fourth Advisor
Pengfei Dong
Abstract
Machine learning, particularly deep neural networks, has demonstrated significant potential in predicting high-dimensional tasks across various domains. This work encompasses a detailed review of Generative AI in healthcare and three studies integrating machine learning with finite element analysis for predicting biomechanical behaviors and properties. Initially, we provide a comprehensive overview of Generative AI applications in healthcare, focusing on Transformers and Denoising Diffusion models and suggesting potential research avenues to address existing challenges.
Subsequently, we addressed soccer-related ocular injuries by combining finite element analysis and machine learning to predict retinal mechanics following a soccer ball hit rapidly. The prediction errors are between 3.03% and 16.40% for peak von Mises stress and maximum principal strain, respectively.
Furthermore, we developed end-to-end deep learning tools as an efficient alternative to finite element methods for predicting stress-strain fields within arterial walls, which is crucial for understanding atherosclerosis. Utilizing U-Net-based convolutional neural networks and conditional generative adversarial networks, combined with ensemble and transfer learning techniques, we developed models that accurately predicted von Mises stress and strain fields in the artery.
Lastly, we developed a finite element-based machine learning model for predicting the mechanical properties of bioglass-collagen composite hydrogels. This model effectively predicted Young’s modulus and Poisson’s ratio with 95% and 83% R-squared values.
This work aims to advance the intersection of machine learning, biomechanics, and healthcare. By developing accurate models for predicting biomechanical behavior and mechanical properties and providing a comprehensive review of Generative AI applications, we hope to contribute to addressing pervasive public health issues.
Recommended Citation
Shokrollahi, Yasin, "Deep Learning and Generative AI Approaches for Automated Diagnosis and Personalized Treatment: Bridging Machine Learning, Medicine, and Biomechanics in Predicting Tissue Mechanics and Biomaterial Properties." (2023). Theses and Dissertations. 1402.
https://repository.fit.edu/etd/1402