Doctoral Defense - Aniwat Juhong

About the Event

Department of Electrical and Computer Engineering

Michigan State University  

Ph.D. Dissertation Defense 

Monday, March 3, 2025, at 12:00 pm   

Zoom Meeting 

Contact Department or Advisor for Zoom Information 

ABSTRACT

MACHINE INTELLIGENCE-ENABLED MULTIMODAL BIOMEDICAL IMAGING

BY: ANIWAT JUHONG   

ADVISOR: Dr. ZHEN QIU

Due to the rapid development of computational technologies, deep-learning-based approaches have emerged as practical and promising remedies for a wide range of biomedical applications. Deep learning can be divided into two main types: supervised and unsupervised learning. Supervised learning aims to define a function that can map input images to their outputs or labels such as classification, segmentation, and regression problems. On the other hand, unsupervised learning defines a function that can extract the latent features and structures from unlabeled data such as clustering problems, dimensional reduction, and super-high-resolution problems. This means we can potentially exploit deep learning to not only extract features and learning pattern from complex data (classification, segmentation, regression, etc.), but also facilitate existing modalities to provide better results in terms quality and acquisition time, such as image resolution enhancement and image denoising. This dissertation demonstrates the utilization of deep learning approaches across multiple modalities in the field of biomedical applications: histopathology image analysis, multispectral optoacoustic tomography (MSOT), magnetic particle imaging (MPI), and Raman spectroscopy. First, we propose custom convolutional neural networks (CNNs) for super-resolution image enhancement from low-resolution images and characterization of nuclei from hematoxylin and eosin (H&E) stained breast cancer histopathological images by using a combination of generator and discriminator networks so-called super-resolution generative adversarial network-based on aggregated residual transformation (SRGAN-ResNeXt) to facilitate cancer diagnosis in low resource settings. Second, we propose deep learning based on hybrid recurrent and convolutional neural networks to generate sequential cross-sectional optoacoustic images. A multispectral optoacoustic tomography (MSOT) system was utilized to acquire the dataset of breast tumors for training our deep learning model. This system can simultaneously acquire the sequential images (cross-sectional images) of MSOT and ultrasound. Furthermore, it provides a spectral unmixing algorithm applied to the MSOT images for extracting the sequential images of a specific exogenous contrast agent. This study used ICG-conjugated superparamagnetic iron oxide nanoworms particles (NWs-ICG) as the contrast agent. Our deep learning model applies to all three modalities (multispectral optoacoustic imaging at a specific wavelength, ultrasound, and NWs-ICG optoacoustic imaging). The generated 2D sequential images were compared to the ground truth 2D sequential images acquired using a small step size showing expressive results.  Third, to perform a high-accurate and noise-invariance deep learning-based approach for MPI-CT image segmentation, we propose the multi-head attention U-Net model, an efficient end-to-end deep learning based on U-Net architecture and multi-head attention mechanism, for MPI-CT image segmentation. The optimal number of attention heads was experimentally observed in this study. Although an increase in the number of attention heads can potentially boost the model’s capability, the excessive number of attention heads results in a decline in capability. Our study shows that the attention U-Net with 4 heads is the most favorable architecture for MPI-CT image segmentation.  Lastly, it is a custom-made Raman spectrometer together with computer vision-based positional tracking and monocular depth estimation using deep learning for the visualization of 2D and 3D surface-enhanced Raman Scattering (SERS) nanoparticles (NPs) imaging, respectively. The SERS NPs used in this study (hyaluronic acid (HA)-conjugated SERS NPs) showed clear tumor targeting capabilities (target CD44 typically overexpressed in tumors) by an ex vivo experiment and immunohistochemistry. The combination of Raman spectroscopy, image processing, and SERS molecular imaging, therefore, offers a robust and feasible potential for clinical applications. 

Persons with disabilities have the right to request and receive reasonable accommodation. Please call the Department of Electrical and Computer Engineering at 355-5066 at least one day prior to the seminar; requests received after this date will be met when possible.