Program

Dr. Chia-Feng Lu, Ph.D., is a board-certified radiological technologist and professor in Taiwan. For the past nine years, he has designed and taught more than fifteen courses related to computer science, programming language, medical imaging techniques, and image analysis for undergraduate and graduate students. Not only in-class teaching, but he is also continuously providing online self-learning materials, including more than 350 YouTube teaching videos that have accumulated over 148 thousand hours of viewing time. He was awarded the Distinguished Teaching Award in the National Yang Ming Chiao Tung University (NYCU) and participated in the innovative teaching program to redesign the course of programming language and implement new teaching technologies to enhance student learning. He works closely with colleagues and students to continuously improve the teaching and learning environment in higher education of medical imaging, including MRI, CT, and fNIRS with computer-aided diagnosis and prognosis. A major research focus recently is to develop image platforms with artificial intelligence in quantifying tumor phenotypes and neurovascular diseases, predicting treatment outcomes using multimodal radiomics (MRI/CT/PET/US), and performing neurological studies of fNIRS, fMRI, and DTI. The issue of patentability and the subsequent technique transfer for the developed techniques and platforms is of interest to him.

Functional Magnetic Resonance Imaging (fMRI) serves as a cornerstone in cognitive neuroscience for mapping brain activity and decoding complex neural networks. This workshop provides a comprehensive overview of blood-oxygen-level-dependent (BOLD) signal analysis using the Statistical Parametric Mapping (SPM12) framework. The curriculum begins with an introduction to fMRI preprocessing, detailing the mandatory spatial and temporal alignment steps required before statistical inference. These foundational techniques include slice timing correction, image realignment, anatomical co-registration, tissue segmentation, spatial normalization, and smoothing.
Building upon these principles, the module shifts to practical application via processing with a real fMRI dataset using SPM. Using real N-Back working memory data, participants are guided through the core stages of fMRI workflows: specifying first-level within-subject models using the General Linear Model (GLM) for parameter estimation, and configuring second-level between-subject designs. Through one-sample and two-sample t-tests, as well as full factorial design configurations, participants will learn to identify consistent brain activations and analyze complex group-by-condition interactions. Ultimately, this framework equips learners to generate robust statistical maps and draw meaningful neuroimaging biomarkers.

Dr. Chien graduated from National Cheng Kung University in 2009 and completed his neurology residency at National Cheng Kung University Hospital in Tainan, Taiwan, from 2010 to 2014. He served as an Attending Physician at the Dou-Liou Branch of the hospital from 2014 to 2016 and has been an Attending Physician at the main hospital since 2016. In parallel, Dr. Chien pursued a PhD in Biomedical Engineering at National Cheng Kung University (2015–2023), focusing on biomedical signal processing, digital image processing, and machine learning. His research earned him the opportunity to deliver an oral presentation at the 2023 MDS Congress, where he discussed the application of machine learning in fMRI studies of Parkinson’s disease. Recently, he was promoted to Assistant Professor in the Department of Neurology at the College of Medicine, National Cheng Kung University (2025). Dr. Chien’s clinical and research interests center on integrating biomedical engineering techniques to enhance the diagnosis and treatment of parkinsonian disorders. His work spans structural and functional neuroimaging studies of movement disorders, the biophysiology of auditory perception, and therapeutic strategies for Parkinson’s disease. He has also contributed to innovative projects focused on digital sensing and cueing devices for Parkinson’s disease, merging biological and digital approaches to advance the understanding and management of neurological disorders.

Transforming imaging features into clinical prediction
This session introduces a practical machine-learning workflow for analyzing cognitive impairment in Parkinson’s disease using resting-state fMRI data. Building on Session 1, participants will use fMRI-derived outputs and parameters, together with the publicly available OpenNeuro dataset ds005892, which is part of a longitudinal study investigating Parkinson’s disease and associated cognitive impairment
The session will demonstrate how neuroimaging features can be extracted and organized for supervised classification. Candidate parameters will include functional connectivity matrices, amplitude of low-frequency fluctuation, and independent component analysis-based features. Participants will learn how these imaging markers may represent brain network alterations related to cognitive dysfunction in Parkinson’s disease.
Through hands-on exercises, participants will construct simple machine-learning pipelines using commonly applied classifiers, including support vector machine, k-nearest neighbors, naïve Bayes, decision tree, and related models. The session will cover essential steps such as feature preparation, model training and testing, performance evaluation, and interpretation of results. Emphasis will be placed on practical implementation rather than advanced mathematical theory. By the end of the session, participants will understand how fMRI-derived matrices and parameters can be transformed into clinically meaningful predictive features for exploring cognitive impairment in Parkinson’s disease.

Po-Yu Lin, M.D., is a clinician-scientist with a specialized focus on neurogenetic and neuroimmunology diseases. He received his medical degree from National Cheng Kung University (NCKU) and completed his resident training in the Department of Neurology at NCKU Hospital. He is currently a Ph.D. candidate in Biomedical Engineering at NCKU and holds a master’s degree in biomedical informatics from the NYU Grossman School of Medicine, where he graduated with the Excellence in Research Award. Driven to bridge the gap between artificial intelligence, clinical neurology and genetics, his research leverages machine learning to build clinical decision support systems. Specifically, he focuses on developing well-calibrated frameworks to integrate diverse evidence and applying genomic language models for precise variant pathogenicity classification.
Beyond his core expertise in genomics, Dr. Lin possesses a solid foundation in applying computational models to diverse clinical data. In this workshop, he draws on his machine learning background to guide a hands-on session on basic deep learning and artificial intelligence-assisted coding. By walking through a practical example of REM sleep behavior detection using polysomnography, he aims to provide neurologists with an accessible, step-by-step introduction to deep learning in neurophysiology.

Artificial intelligence and machine learning are increasingly applied to clinical neurophysiology, yet the technical barrier to entry remains high for most clinicians. This workshop addresses that gap through a fully interactive, hands-on experience requiring no prior programming background.
Using polysomnography (PSG) as the primary signal modality, participants will be guided through a completely deep learning pipeline: from raw neurophysiological signals to clinically interpretable predictions. The session introduces key concepts in biomedical time-series analysis and deep learning, with each concept immediately reinforced through live coding exercises.
A central feature of this workshop is the use of AI-assisted coding, allowing participants to build and modify machine learning models interactively with the help of generative AI tools. This approach demonstrates how modern AI can lower the technical barrier and accelerate model development in clinical research settings.
The application focus is the automated detection of REM sleep behavior disorder (RBD), one of the most powerful prodromal markers of Parkinson's disease and related synucleinopathies. By the end of the session, participants will have hands-on experience with a real-world deep learning pipeline, and a practical understanding of how these methods can be translated to clinical neurophysiology research.