| Proj No. | A1039-251 |
| Title | Deep Learning-Based Surgical Tool Detection and Segmentation for Minimally Invasive Surgery |
| Summary | Accurate detection and segmentation of surgical instruments are critical for enhancing safety, precision, and automation in minimally invasive surgery. This project aims to design, train, and evaluate deep learning models capable of identifying surgical tools in endoscopic images, forming a foundation for intelligent computer-assisted surgery systems. The project begins by studying existing deep learning frameworks in visual recognition tasks to understand core principles such as heatmap regression, encoder–decoder architectures, and dataset augmentation. These insights are then applied to adapt and implement models suitable for surgical environments, addressing challenges like occlusion, varying illumination, and complex anatomical backgrounds. Models will be trained and tested using publicly available surgical video datasets, with performance evaluated through standard detection and segmentation metrics. Iterative experimentation will explore architectural variations and fine-tuning strategies to improve accuracy and robustness. The results are expected to advance reliable visual recognition in surgical applications, supporting real-time guidance and future integration with robotic-assisted systems. |
| Supervisor | Ast/P Huang Hen-Wei (Loc:S2 > S2 B2A > S2 B2A 08, Ext: ?) |
| Co-Supervisor | - |
| RI Co-Supervisor | - |
| Lab | Schaeffler Hub for Advanced Research (SHARE) at NTU (Loc: S2.1-B4-01a) |
| Single/Group: | Single |
| Area: | Intelligent Systems and Control Engineering |
| ISP/RI/SMP/SCP?: |