Degree
Doctor of Philosophy (PhD)
Department
Department of Electrical and Computer Engineering
Document Type
Dissertation
Abstract
Modern imaging techniques, such as X-ray imaging, magnetic resonance imaging (MRI), and ultrasound, play important roles in disease diagnosis nowadays. However, their disadvantages, such as ionizing radiation, low resolution, and unsatisfactory accuracy, may limit their applications. Innovative biomedical techniques are necessary for better head and neck health disorder precautions and treatment. In the dental imaging project, we designed and developed an Indocyanine Green (ICG) assisted near-infrared (NIR) fluorescence (NIRF) dental imaging system for dental diseases (caries, cracks, and rebuilt crown) examination and diagnosis in a human extracted tooth model. Excellent image contrast was illustrated in short (1 minute) imaging windows (time duration from ICG solution immersing to imaging) in both NIR-I (wavelength between 700 nm - 950 nm) and NIR-II (1000 nm - 1700 nm); cracks were located precisely while micro-computed tomography (CT) failed to recognize. In addition, tooth images with cracks and without cracks were extracted and collected from video frames for further analysis with multiple Deep Learning (DL) approaches. Rebuilt teeth, including the decayed human tooth root and dental crown, were also examined with additional long-wavelength light sources for practical rebuilt tooth decay detection. 1650 nm LED outperformed other wavelength LEDs, demonstrating the most distinct decay outline of the root. In the Machine Learning (ML) and Raman spectroscopy-aided laryngeal cancer identification project, we collected both human and mouse cancerous and non-cancerous Raman spectra; different classification models such as Random Forest (RF) and one dimensional (1D) convolutional neural networks (CNN) were applied the datasets. The 1D-CNN human model showed an average accuracy of 96.1%, a sensitivity of 95.2%, and a specificity of 96.9%; the 1D CNN mouse model illustrated a better performance than the human model, with an overall accuracy of 99.8%, a sensitivity of 99.7%, and a specificity of 99%. Overall, the novel xiv ICG-assisted mouthwash NIRF dental imaging method serves as an ionizing-free and effective tool for dental disease examination; the combination of ML and Raman spectroscopy has the potential to be an intraoperative device for laryngeal cancer identification.
Date
10-31-2023
Recommended Citation
Li, Zheng, "NEAR-INFRARED IMAGING FOR HEAD AND NECK HEALTH: DENTAL DISORDER AND LARYNGEAL CANCER DIAGNOSIS" (2023). LSU Doctoral Dissertations. 6290.
https://repository.lsu.edu/gradschool_dissertations/6290
Committee Chair
Xu, Jian