An integrated research program centered on brain tumor stereotactic radiosurgery. It features an advanced pipeline combining lesion-sensitive segmentation, quantitative MRI analysis, and constrained VLM-based report generation—all enabled on non-contrast MRI via knowledge distillation.

• Utilizes knowledge distillation to successfully detect and segment brain tumor lesions directly from non-contrast MRI scans.
• Integrates non-contrast MRI lesion detection into a constrained vision-language system to automate clinically grounded report generation.
• Encompasses core Ph.D. work on DBI-MambaUNet and SA-FTL while standardizing outputs into PACS-compatible DICOM SEG/SR.

A dedicated method-development project for multi-subtype brain tumor micro-lesion analysis in stereotactic radiosurgery, with explicit focus on improving lesion sensitivity for small lesions.

• Core manuscript currently under review at Medical Image Analysis.
• Designed to improve sensitivity for sub-4 mm lesions while maintaining clinical relevance.
• Represents the most method-centric part of the current Ph.D. work.

A clinically oriented brain tumor AI product successfully transitioning from research to clinical validation. It has achieved both Taiwan FDA (TFDA) and U.S. FDA (510(k)) clearance. It is now officially deployed and in actual clinical use at Taipei Veterans General Hospital, Taichung Veterans General Hospital, and Shin Kong Wu Ho-Su Memorial Hospital.

• Successfully advanced the system through rigorous TFDA certification pipelines.
• Led the product integration and deployment across multiple top-tier medical centers in Taiwan.
• Bridged the gap between advanced medical AI research and real-world clinical applications.

An advanced brain tumor AI product driving the next generation of diagnostics. Strongly positioned for both Taiwan FDA (TFDA) and U.S. FDA medical device applications. Like its predecessor, this system has already been deployed for clinical use at Taipei Veterans General Hospital, Taichung Veterans General Hospital, and Shin Kong Wu Ho-Su Memorial Hospital.

• Accelerating regulatory compliance and application processes for upcoming TFDA and U.S. FDA approvals.
• Achieved rapid clinical deployment and validation in major medical center workflows.
• Demonstrates independent technical leadership and strong capabilities in establishing robust, product-facing medical AI.

An AI-assisted rehabilitation platform designed for home and clinical environments, developed under CTO leadership with a focus on functional assessment, workflow design, and translational deployment.

• Led technical strategy, system architecture, AI workflow design, and prototype development.
• Selected as a Success Case on Taiwan Ministry of Education's College Innovation and Entrepreneurship Simulation Learning Platform.
• Advanced through FITI and continued into entrepreneurship-oriented development programs.

A 3D CT segmentation framework for acute appendicitis built for the AOCR Kaggle AI Challenge, emphasizing lesion-sensitive detection, localization, and clinically robust performance.

• Led the research team across model design, optimization, and evaluation.
• Won the championship in a challenge organized under AOCR, one of the three major international radiology conferences.
• Generated downstream impact including awards, patent protection, and media visibility.

A healthcare-oriented student project focused on a smart medication delivery robot, combining computer vision, sensor fusion, autonomous navigation, and practical system design thinking.

• Mentored students on technical framing, presentation logic, and healthcare application planning.
• Supported system design discussions and evaluation planning.
• Highlighted safe, practical, and deployment-oriented AI thinking in student work.

A mentoring project on medication recognition and grounded medical education using image detection/classification, retrieval-augmented generation, and large language models.

• Guided system architecture design for image-based medication recognition.
• Emphasized grounded responses, citation-aware output, and hallucination risk reduction.
• Strengthened the practical educational value of the project in healthcare settings.

An interactive generative imaging system for the National Museum of Natural Science that used controllable diffusion workflows and exhibit-specific adaptation to improve realism and user interaction.

• Implemented ControlNet and LoRA for controllable, exhibit-aware generation.
• Integrated diffusion-based generation and face-swapping techniques to improve realism.
• Received the Excellence Award in the AI Application Challenge hosted by Taiwan’s Ministry of Digital Affairs.

A cGAN-based synthetic data and cross-modal translation project addressing data scarcity and hardware-constrained deployment in drone vision scenarios.

• Built synthetic image generation workflows for data expansion and downstream detection support.
• Worked on RGB-to-infrared translation to strengthen cross-modal perception in restricted environments.
• Focused on practical deployment constraints for aerial computer vision systems.