Imaging plays an important role in all clinical processes. One challenge in medical image data processing is detection and tracking objects and instruments, which faces complications arising from the developed medical image acquisition systems and also the nature of in-vivo medical images. Special properties of the in-vivo bio images such as noise, specular highlights, inhomogeneity, heterogeneity, varying luminosity, and background change, in addition to the changes of camera, out of camera view tools, and multiple moving tools (instrument tools, surgical suture, cutting instrument, tissue movement) make object detection and tracking in the biomedical image processing complicated. In this study, the k-means clustering method in combination with the level set active curve model are used to develop a platform for low-cost tracking of surgical tools in robotic surgery videos. After removing the image background, the smoothed image is used as input to the numerical method. This model tracks the robot tools even when the camera view changes, the tool is lost, the tissue is bleeding and moving, and the luminosity of the images changes. The developed model is validated using video frames of real and simulated robotic surgeries. The accuracy of model in tracking da vinci robot end-effectors for a video with 12000 frames, recorded at Roswell Park Cancer Institute, is 93%. Accuracy of proposed framework is compared to those for existing numerical models, DRLSE and Chan-Vese. The results show that proposed surgical robot tool tracking model is more efficient than existing computational models.

This content is only available via PDF.
You do not currently have access to this content.