Probabilistic neural network applied to eye tracking control to alter the direction of an endoscopic manipulator
In this study, we propose a novel endoscopic manipulation system that is controlled by a surgeon's eye movements. The optical axis direction of the endoscopic manipulator is altered intuitively based on the surgeon's pupil movements. A graphical user interface was developed by divi...
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
The Japan Society of Mechanical Engineers
2017
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/a70db1c778b04fbdac1e0ffdf463ae03 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | In this study, we propose a novel endoscopic manipulation system that is controlled by a surgeon's eye movements. The optical axis direction of the endoscopic manipulator is altered intuitively based on the surgeon's pupil movements. A graphical user interface was developed by dividing the monitor screen into several areas with shape boundaries, so that the movement direction of the endoscope can be identified by the area gazed at by the operator. We used a probabilistic neural network (PNN) to decide the regional distribution proportion to recognize the direction in which the operator would want the endoscopic manipulator to move. The PNN model was trained by individual calibration data. We hypothesized that PNN model training could be completed immediately after calibration, which also determines the boundary of the regional distribution portion (RDP). We designed an experiment and recorded the path of direction change to verify the PNN's effectiveness in our proposed system. All participants, including four who wore glasses, completed the requested task. Moreover, wearing glasses had no significant effect on the performance of the proposed system. Furthermore, the PNN training duration only took 2% of the entire time of the procedure to handle individual differences. We conclude that our method can handle individual differences in operators' eyes through machine learning of personal calibration data over a short time frame, which will not take significant extra preoperative setup time. |
|---|