Session Index

As the final layer of the central dogma, metabolomes directly reflect phenotypic changes in human bodies. However, the cumbersome sample preparation process for mass spectroscopic measurement and the lack of spatial information in tissues limit their wide application in clinical practice. Here, we propose the use of fluorescent metabolomics, which consists of metabolites that can emit fluorescence in blood, saliva, and urine. In comparison to conventional mass spectroscopic measurement, fluorescent metabolomics allows for direct measurement without the need for reagents. It can serve as a continuous monitoring device for detecting the emergence of acute illnesses. Furthermore, when combined with a multiphoton imaging system, spatial metabolomic information can be obtained with sub-cellular resolution. In this lecture, we will demonstrate the use of fluorescent metabolomics in evaluating organoid pharmacokinetics, detecting the presence of acute mesenteric ischemia, and aiding in the differential diagnosis of prediabetes. This provides a new avenue toward precision medicine.

  Preview abstract

With a miniaturized scan lens and a micro-electro-mechanical system (MEMS) mirror, we designed the 1st-gen miniaturized endoscopy system which is called “NTU miniScope”. Through different designs of scan lens and gradient refractive index (GRIN) lens pair, one can easily substitute the imaging head of the NTU miniScope by using a magnetic-base design. Moreover, with a sparse spiral scanning pattern, a 600Hz frame rate can be realized for voltage imaging.

  Preview abstract

This study investigates enhancing endoscopy image quality through deep learning. We explore supervised and unsupervised approaches using a dataset of 3500 images. Preprocessing techniques reduce noise, and experiments highlight deep learning’s potential. The study aims to improve image quality and elucidate effective methodologies.


  Preview abstract

Endo-microscopy with structured illumination HiLo imaging process can provide sectioning ability. However, moving optical elements in the axial direction for the 3D imaging restricts compact endoscopic system design. In addition, HiLo imaging requires multiple shots, which is time-consuming and make system configuration complicated. Here, we propose an in-vivo intelligent fluorescence sectioning using meta-varifocal endo-microscopy. With the telecentric design, the endo-microscopy can provide constant magnification during axial scanning for in-vivo 3D imaging of mouse brains. Furthermore, we introduce the deep learning (DL) network for HiLo sectioning technique, which can substantially reduce image acquisition time and system complexity.

  Preview abstract

Understanding human brain function is crucial yet challenging due to its complexity, especially in Attention-Deficit/Hyperactivity Disorder (ADHD). This study addresses the issue by utilizing cerebral organoids from human pluripotent stem cells (hiPSCs) and employing a custom two-photon microscopy and Fluo-4 indicator for calcium imaging and neural activity analysis. To handle the extensive and complex streaming data, we propose an analytics methodology. Results exhibit synchronization patterns and frequency-amplitude differences between different genotypes. Our analysis not only provides comprehensive insights into neuron dynamics but also holds promise for understanding the intricacies of brain function and dysfunction.

  Preview abstract

Mammograms are critical for early detection of breast cancer. However, the accuracy of mammogram readings varies and depends on radiologists' experience and skill. Artificial intelligence (AI), with its visualization capability, can assist physicians in detecting cancer. However, current AI models often detect lesions but lack information on types and the Breast Imaging Reporting & Data System (BI-RADS) scores. In this study, we propose an approach that leverages mammogram resources, biopsy results, and the BI-RADS scoring procedure of radiologists to predict breast malignancies on our private dataset of 155 patients using our custom model.

  Preview abstract

Swept source optical coherence tomography (SS-OCT) is a high-speed, and non-contact imaging modality that can provide three-dimensional tissue structure. With the advantage of long coherence length, MEMS tunable vertical cavity surface emitting laser (VCSEL) is a suitable light source for full-eye imaging. In this work, we have developed a full-eye imaging engine by using a novel wavelength-swept laser light source based on high-contrast grating (HCG) VCSEL. Full eye model imaging and the axial length measurement as well as the evaluation of the tuning noise of the HCG-VCSEL are performed at an A-scan rate of 30 kHz.

  Preview abstract