Highlights of the research by the professor of Biomedical Electronics and Bioinformatics Institute. -202209 Associate Professor Sung, Kung-Bin

With the improvement of computation technologies, algorithms have become one of emerging research fields. Professor K.-B. Sung uses different algorithms to solve the quantification and interpretation of various biomedical tests, thereby improving the accuracy of medical diagnosis.

First of all, Prof. Sung published a work in the Journal of Biomedical Optics in April at this year- “Characterization and identification of cell death dynamics by quantitative phase imaging.” During drug development, in this work, the progression of cell death, e.g. the generation of extracellular vesicles during apoptosis, has been investigated for a comprehensive understanding of cell responses with Quantitative phase imaging (QPI). QPI is a label-free imaging technique that records the phase distribution of a cell resulting from both the intracellular refractive index and cell thickness. This study applies QPI to classify apoptotic, necrotic, and normal cells simultaneously. It characterizes the dynamics of morphological and quantitative-phase features in the cell death process. In conclusion, this study showed 84.0% overall accuracy in classifying normal cells.

Prof. Sung has also published another works in Frontiers in Physics in June of this year-“Non-Invasive Quantification of Layer-Specific Intrinsic Fluorescence From Mucosa of the Uterine Cervix Using Monte-Carlo-Based Models.” Previously, many in-vivo studies showed differences in the intensity and/or shape of fluorescence emission from the mucosa in more accessible organ sites such as in the digestive tract and urogenital system. These may provide helpful information for the early detection of epithelial cancers or their precursors. And major fluorophores in the mucosa that are related to cancerization include reduced nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD) found predominately in mitochondria of cells. Using Monte Carlo (MC) simulations of fluorescence excitation and emission processes has advantages of being flexible and versatile because they can accommodate geometric specifications of any experimental arrangement and tissue heterogeneity. In this study, Prof. Sung combined MC simulations and artificial neural networks (ANNs) to reduce processing time. Regardless of the sensitivity of detecting cervical precancers, using intrinsic fluorescence alone may not be sufficient. The time required for extracting tissue’s optical properties could be accelerated by a fluorescence ANN forward model and optimizing the processing speed. Moreover, spectra captured from scanning a large tissue area could be analyzed simultaneously through parallel operations to develop this method as an instant diagnostic tool.

Finally, Prof. Sung published a wrok in the Journal of Biomedical Optics in August this year- ” Quantifying tissue’s optical properties of human heads in vivo using continuous-wave near-infrared spectroscopy and subject-specific three-dimensional Monte Carlo models.” Human brain activities have long been of great interest to many fields, stretching from fundamental neuroscience, branches of medicine, and education to even the brain-computer interface. Functional magnetic resonance imaging (fMRI) is the standard gold method of brain activity monitoring with great three-dimensional (3D) spatial resolution over the whole brain. However, the temporal resolution of fMRI is low, and the subject or patient is restricted to the supine position in an MRI scanner, which is expensive and not readily accessible. Functional near-infrared spectroscopy (fNIRS) employs low-cost and compact instruments to monitor brain activities with moderate temporal and spatial resolution. This study aimed to use a multidistance continuous-wave (CW) NIRS system to quantify the scalp, skull, and GM’s (gray matter) optical properties (Ops) in the human head in-vivo. In the end, the average error between ANN outputs and corresponding MC simulations was under 10%! Moreover, the curve fitting could be done in several minutes with the help of the ANNs.

In conclusion, the improvement of computing has further promoted the progress of science at various levels. I believe that the future of medicine is just around the corner!



  1. Huai-Ching Hsieh, Po-Ting Lin, Kung-Bin Sung, “Characterization and identification of cell death dynamics by quantitative phase imaging,” J. Biomed. Opt. 27(4) 046502 (28 April 2022)https://doi.org/10.1117/1.JBO.27.4.046502
  2. Lin G-S, Tu S-C, Mok C-I, Huang T-H, Chen C-H, Wei L-H and Sung K-B (2022) Non-Invasive Quantification of Layer-Specific Intrinsic Fluorescence From Mucosa of the Uterine Cervix Using Monte-Carlo-Based Models.  Phys.10:865421. doi: 10.3389/fphy.2022.865421
  3. Tzu-Chia Kao, Kung-Bin Sung, “Quantifying tissue optical properties of human heads in vivousing continuous-wave near-infrared spectroscopy and subject-specific three-dimensional Monte Carlo models,” J. Biomed. Opt. 27(8) 083021 (22 June 2022) https://doi.org/10.1117/1.JBO.27.8.083021





首先介紹宋老師在今年四月刊登在Journal of Biomedical Optics的文章-”利用定量相位顯示技術表徵及判斷細胞死亡動態”(1.)。在藥物開發的時候,需要經過不同的篩選,過程中會利用螢光染劑來觀測不同種類的細胞死亡,以此來獲得藥物篩選所需的重要資訊。而此篇研究中所利用的定量相位顯示技術(Quantitative Phase Imaging,QPI)便是一種免標記且可以快速觀測的方法。利用光學的攝像技術加上電腦程式的輔助分析,可以實現細胞凋零的分類器。藉由分析細胞的不同參數(如:細胞圓形性、離心率、光學體積等…)最後得到分類結果的最佳表現。在文中表明,利用QPI的技術可以達到總體平均高達84%的準確度!

再來要介紹的是在今年六月刊登在Frontiers in Physics的文章-“利用蒙地卡羅模型來實現非侵入式定量子宮頸黏膜組織內在螢光特徵” (2.)。子宮頸癌一直以來都是女性癌症的主因之一,而根據研究顯示,在癌症逐漸形成的同時,其中的NDHD(一種負責人體細胞能量代謝的功能蛋白)比例會逐漸升高。而藉由定量黏膜組織中的螢光物質特徵,有助於定量組織中的NDHD。而宋老師的團隊致力於利用光譜系統來進行活體的子宮頸漫反射光譜以及螢光物質定量,利用光譜系統的好處在於避免了傳統利用組織切片的侵入式手段,並且能夠更全面的檢測目標組織。為了完成光譜的擬合,利用了蒙地卡羅模型(Monte Carlo method,MC)來進行分析,而宋老師的團隊更引入了人工神經網路(ANN)來加速擬合的速度,雖然在文末指出這樣的方式準確度也許不盡理想,但是藉由這樣的模擬分析仍然很有潛力,是可以持續開發優化的目標!

最後在今年八月一樣刊登在Journal of Biomedical Optics的文章-“基於蒙地卡羅模型來進行人體腦部光學參數定量的臨床實驗,利用連續近紅外光譜與三維模型” (3.)。在目前的人體大腦研究,都是利用功能性磁振造影(fMRI)或是腦電圖(EEG)來做腦部造影,但前者不意分辨收到信號的前後時間;後者不易分辨收到信號的來源深淺。所以宋老師的團隊發展了利用光學進行量測的近紅外光譜(NIRS),由於其光學的特性,可以達到比EEG還來的更高空間解析度並且同樣具有設備小巧可攜的特性,是有望取代EEG的技術。而基於MC模擬以及ANN的結合,此研究可以達到10%以下的誤差,而且將檢測時間也僅需數分鐘即可。