TY  - UNPB
AV  - public
M1  - Doctoral
Y1  - 2024/06/28/
TI  - Surgical Gesture Recognition in Robot-Assisted Laparoscopic Surgery
EP  - 152
UR  - https://discovery.ucl.ac.uk/id/eprint/10193373/
ID  - discovery10193373
N2  - Automated surgical gesture recognition is a crucial step in developing systems for surgical data science, analytics, objective skill evaluation, and surgical automation. However, it is a challenging task due to the high variability of surgical gestures caused by various factors such as the surgeon's style and the patient's anatomy. This variability leads to differences in the duration, kinematics, and order of actions across different demonstrations, making the analysis and understanding of surgical instrument kinematics and video prone to errors. Another challenge is the lack of large and diverse datasets with annotated demonstrations, which are essential for training robust recognition systems based on machine learning and deep learning. Manual labeling of surgical gestures is costly, time-consuming, and prone to errors and inconsistencies. The goal of this thesis is to propose new strategies for integrating complementary information, such as different tasks or input modalities, to improve recognition performance and reduce the reliance on large training datasets. The thesis first presents a multi-task network for surgeme recognition and progress prediction, which achieves improved performance without the need for additional manual labeling and training. The thesis also focuses on data fusion for action recognition. It explores methods to mitigate prediction errors caused by individual sensor noise and limitations by dynamically weighting each modality over time. Additionally, it investigates techniques for enhancing multimodal representation learning by separating hidden features into modality-shared and private components. In both cases, the experiments demonstrate performance improvements over unimodal baselines and common multimodal fusion strategies. The evaluation is conducted on established benchmark datasets and a newly developed dataset of real surgical interventions. By addressing the challenges of surgical gesture recognition through the integration of complementary information and the development of new datasets, this thesis contributes to advancing the field and enhancing the applicability of recognition systems in real surgical settings.
N1  - Copyright © The Author 2024. Original content in this thesis is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) Licence (https://creativecommons.org/licenses/by-nc/4.0/). Any third-party copyright material present remains the property of its respective owner(s) and is licensed under its existing terms. Access may initially be restricted at the author?s request.
A1  - van Amsterdam, Beatrice
PB  - UCL (University College London)
ER  -