01
Education
2024 - 2026
National Cheng Kung University
Institute of Computer Science and Information Engineering
2020 - 2024
University of Taipei
Department of Computer Science
02
Relevant Projects
Intelligent Pressure Sensing Mattress System
Master's Lab Project
Python
PyQt6
YOLOv11 (Deep Learning)
FT4222 (IoT)
OpenCV
Integrated software and hardware for real-time pressure monitoring using a vendor-provided 27x72 Pressure Sensor Matrix.
- Implemented Real-time Data Acquisition via FT4222 USB-SPI Bridge to stream raw pressure heatmaps.
- Deployed YOLOv11 for posture recognition (Supine, Lateral, Out-of-Bed) and applied Dynamic Calibration for noise reduction.
- Developed a PyQt6 Application for visualization and automated reporting.
Multimodal In-Vehicle Smart Assistant
Internship Project @ CarUX
Python
PyTorch
Llama 3
AVSR
PyQt5
Developed a user intent analysis system solving Voice Activity Detection (VAD) failures in noisy environments through Multimodal Fusion.
- Implemented Lip-Reading VAD (Visual Speech Recognition) to complement audio signals in high-noise cockpit scenarios.
- Integrated Llama 3 LLM to process consolidated Audio-Visual inputs for precise vehicle command execution.
- Designed the integration pipeline for open-source speech models (AVSR) within a real-time PyQt5 system.
Go Game Strength & Style Recognition
AI Cup 2023 (Rank 18/580)
Deep Learning
RestNet
Ensemble Learning
Designed a deep learning architecture inspired by AlphaGo to estimate player rank strength (Dan/Kyu).
- Constructed a model using Residual Blocks to process 19x19x5 Feature Matrices (Board state, Liberties, History).
- Applied a Weighted Average Ensemble strategy to combine multiple predictors, optimizing classification performance.
- Analyzed playstyle patterns using Policy/Value head concepts.
Multimodal Pathological Voice Classification
Graduation Project
Audio Processing
CRNN
Multimodal Fusion
Developed a diagnostic classification system using Feature Engineering and Model Fusion.
- Utilized CRNN (CNN+LSTM) models to analyze Mel-spectrogram time-series features.
- Combined audio features with clinical metadata through Multimodal Fusion techniques.
- Achieved Rank 57/320 in the AI Cup competition.
03
Journal Publications
Human-Computer Vision Collaborative Measurement of Surgical Exposure and Length in Endonasal Endoscopic Skull Base Surgery
IEEE Open Journal of Engineering in Medicine and Biology, Volume 6 | Impact Factor: 3.3 (2025)
Second Author
Developed a computer vision system for quantitative analysis of surgical exposure in skull base surgery, enabling measurement from endoscopic videos with clinical validation against neuronavigation systems.
95-96%Accuracy
89%Time Reduction
50Videos Analyzed
Early Withdrawal of Systemic Corticosteroid Following Malignant Glioma Craniotomy: A Real-World Population-Based Cohort Study
Journal of Neuro-Oncology
Second Author
Clinical Significance: Investigates the optimal timing and safety of early corticosteroid withdrawal in patients following malignant glioma resection, using real-world population-based cohort data to assess outcomes and complications.
Computer Vision Prediction of Normal Pituitary Gland Lateralization in Pituitary Macroadenoma
Scientific Reports (Nature Portfolio)
Second Author
Developed a computer vision system to identify and determine the lateralization of the normal pituitary gland from preoperative MRI, addressing the clinical challenge where successful preoperative identification was achievable in only 20-60% of patients.
90-96%Accuracy
52%Time Reduction
97Patients
Impact of Preoperative Smoking on Outcomes of Endoscopic Transsphenoidal Surgery for Pituitary Tumor
Endocrine
Co-Author (8th)
Clinical Significance: Investigates how smoking affects wound healing, postoperative infection rates, CSF leak risk, and endocrine function recovery following transsphenoidal surgery.
The Impact of Preoperative Rhinitis on Outcomes of Endonasal Endoscopic Surgery for Pituitary Tumors
Endocrine
Co-Author (3rd)
Clinical Significance: Examines how preoperative rhinitis and nasal mucosal inflammation affect surgical outcomes, postoperative complications (CSF leak, infection), and recovery following endonasal endoscopic pituitary surgery.
04
Conference Publications
Mitigating Age-Dependent Bias in Parkinson's Disease Classification Using a Stacked Ensemble Architecture
IEEE International Conference on Fuzzy Systems (iFUZZY), November 5-8, 2025
Second Author
Proposed a stacked ensemble learning architecture to address age-dependent performance bias in voice-based Parkinson's disease diagnosis. Traditional single-classifier models exhibit high sensitivity but lower specificity in older groups.
0.857Sensitivity
0.871Specificity
Robust and Fair Parkinson's Disease Detection: A Stacked Ensemble Learning Approach Across Age Groups
The 3rd Asia-Pacific BT&D2 Alliance (APBA) International Conference, Tainan, November 28-30, 2025
Poster presentation extending our iFUZZY conference paper with additional visualizations and methodological details for the research community at a major international biomedical conference.