Singularity-free localization of a magnetic medical capsule for colonoscopy

Colonoscopy is a procedure for the detection of anomalies in the terminal part of the human intestine. There are several drawbacks associated with the procedure, as the discomfort felt by the patient or the damage of the tissue walls caused by the friction of the stiff medical instrumentation. These aspects prevent many people from undergoing the test of colonoscopy, thus putting at risk many lives. For this reason, during the last decade new technologies involving robotics have been studied to replace the traditional methods used today. One of the most promising approaches for robotic colonoscopy involves an endoscopic capsule magnetically coupled to a permanent magnet fixed to the end-effector of a robotic arm, whose position is decided by the physician. Due to the difficulties that result from indirectly controlling the pose of the capsule inside the body by moving the robot, a more straightforward solution has been thought, involving a closed-loop system that would allow the physician to use a joystick to directly control the movements of the capsule. This kind of system requires the knowledge of the capsule's pose instant by instant, so a thorough localization method is essential. State-of-the-art techniques use data acquired through sensors mounted on-board of the capsule. Inertial measurements are exploited for the computation of the capsule's orientation. The information of the magnetic field involved in the actuation of the probe is used to find the capsule's position, through the comparison of the measured values with a mathematical model of the magnetic field. Issues are encountered in the position detection due to the symmetrical shape of the magnetic field generated by the permanent magnet: when the capsule assumes certain positions around it, the algorithm that provides the localization encounters a singularity, returning an undetermined number of solutions. In addition to this, these techniques require an initialization of the capsule's orientation to have an overall good estimate of the pose. The work described in this thesis aims at solving the problems related to the undetermined solution in the position detection and in the initialization of the capsule's orientation. The whole work is based on a system for robotic colonoscopy under development at the two divisions of the STORM Lab at Vanderbilt University, Nashville, TN and at the University of Leeds, England, UK. The problem related to the position detection is addressed by adding a second magnetic field to the one generated by the permanent magnet, so that the singularity problem is constrained and only one solution is obtained from the localization. For this to happen, the additional field needs to have a dipole moment orthogonal to the first one and has to be discernible from the first magnetic field. This additional signal is thus generated in the alternating current (AC) domain, so that it can be filtered out from the data acquired through the sensors. The generation of the AC magnetic field is accomplished by means of a coil specially designed for the device. Specific requirements were considered in the coil design about the dimensionality of the object, the heat produced by Joule effect and the minimum magnitude of the generated field at a given distance from the center of the coil. In order to fulfill all the technical requirements, a problem of optimization was implemented for the search of the most appropriate values for the coil building parameters, such as the diameter, the length and the number of wire turns. A custom device was also built and programmed for the excitation of the coil and the acquisition and processing of the data regarding the magnetic field sensed by the capsule. Such device is the combination of an STM32 Nucleo programmable board and a customized electrical circuit. The addition of a second magnetic field involves also the update of the model for the computation of the field in the localization algorithm. Due to the alternating nature of this signal, a solution was implemented to transform the time-varying field into a static one, thus considering the coil as a second permanent magnet; in this way, its values can be computed using the same model exploited for the magnet and added to the original computation of the field. A smart processing method for the collected data (Goertzel Algorithm), implemented in the programmable board, is thus presented: the method consists in a frequency-selective filter for the magnitude detection of the AC magnetic field in correspondence of its main frequency of oscillation. The entire system was validated through four experimental tests. In the first place the effects of the overlap of two fields of different nature were studied: the aim of the test was to verify the validity of the superimposition principle and the eventual demagnetization of the permanent magnet due to an AC current flowing around it. A second test studied the variations in the sensors' readings in case of the presence of material, such as water, between the source of magnetic field and the sensors. In the end, two validations were made to verify that the problems of the singularity encountered by previous methods were solved. Both static and dynamic conditions were considered for the capsule and the robot end-effector. The first two experimental validations proved the possibility to exploit static and time-varying magnetic fields in the desired implementation of a localization method. The static and the dynamic tests allowed instead to prove the correct working of the localization where previous methods failed, accomplishing a pose detection with error below 10mm in position and 6 degrees in orientation. These values are very close to the errors associated to other localization methods and they stay consistent even in presence of singularity. Further improvements are still possible for the system at the level of the programmable board: even though the results from the Goertzel Algorithm are satisfactory, a more precise solution can still be obtained from the filter by either increasing the amount of samples in input or combining the filter with a windowing method that would help avoiding the detection of unwanted frequencies of the signal spectrum.