Localization of a Magnetically Actuated Capsule Endoscope: Performance Assessment and Proposal of an Improved Algorithm

Colorectal cancer is a significant global health concern, ranking as the third most prevalent malignancy and the second most deadly cancer. Early detection of cancer plays a pivotal role in improving the chances of survival. Consequently, through colonoscopy, medical professionals conduct visual examinations of the colon to identify any early signs of cancer. Existing endoscopes have been associated with concerns regarding tissue damage and patient discomfort, leading to a reluctance among patients to undergo recommended screening procedures. For this reason, significant efforts have been made over the past two decades to develop alternative devices. In particular, the STORM Lab has implemented a robotic platform called Magnetic Flexible Endoscopy (MFE). The MFE stands out for its remarkable front-pull actuation of the endoscopic tip facilitated by an external magnet. This revolutionary approach eliminates the need for rear-push mechanical actuation and the use of semi-rigid insertion tubes. Accurately estimating the capsule pose is crucial for magnetic actuation systems to apply the required forces and torques effectively. Therefore, the STORM Lab developed a localization algorithm based on the Particle Filter (PF). The first objective of the thesis project was to estimate in real-time the correctness of the localization algorithm developed at STORM Lab for the MFE. Therefore, a comprehensive exploration of various parameters was undertaken. Initially, tests were conducted to ascertain the interrelationships between the parameters. This process aimed to establish how these parameters should be interconnected and how their values could be combined to yield meaningful insights into the localization quality. Additionally, specific thresholds were defined to discern between good and bad localization. Subsequently, a validation phase was implemented to rigorously examine and confirm the effectiveness of the identified parameters and thresholds in different scenarios. The second goal of the project was to develop a new localization algorithm. For this purpose, two possible algorithms were analyzed, and the Unscented Kalman Filter (UKF) was identified as the most suitable. Following the development of the novel localization algorithm, the UKF was tested to determine its parameters optimally. Subsequently, a series of static tests were conducted to validate the new algorithm and compare its results with those of the PF. The concluding phase of the project involved the fusion of the two developed algorithms, PF and UKF, strategically extracting the strengths of each in pursuit of a unified localization algorithm. Given that the errors resulting from both localization algorithms, PF and UKF, are comparable, it is possible to state that the lower limit of error has been attained and is fundamentally contingent on the system’s intrinsic characteristics. In pursuit of refining localization, any attempts to achieve further enhancements would necessitate altering the system.