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. 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), computed tomography (CT), magnetic particle imaging (MPI), and Raman spectroscopy. The first deep learning application is convolutional neural networks (CNNs) for resolution enhancement and nuclei segmentation of hematoxylin and eosin (H&E) images. This deep learning-based approach could facilitate cancer diagnosis using the H&E images in a low resource setting. The second application is based on hybrid recurrent and convolutional neural networks to generate sequential cross-sectional MSOT images in order to reduce the acquisition time. Essentially, the proposed deep learning model can generate the missing sequential MSOT images in the data acquired by a large step size setting, resulting in a comparable resolution to the data acquired by a small step size setting. The third application is an efficient end-to-end deep learning model based on U-Net architecture and a multi-head attention mechanism for MPI-CT image segmentation. This proposed model can directly segment the MPI signal from the co-registered MPI-CT image with promising performance. 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 combination of Raman spectroscopy, image processing, deep learning, and SERS molecular imaging shows the robust and feasible potential for clinical applications.
Read
