Date: 14th October 2020
Colorectal cancer is the third most commonly occurring cancer, accounting for around 1.8 million new cases in 2018. Early diagnosis substantially improves survival however, excess demand for colonoscopies creates bottlenecks in the diagnosis process resulting in over half of cases receiving a late diagnosis. Now scientists, have used machine vision to develop intelligent and semi-autonomous control of a magnetic endoscope, enabling non-expert users to effectively perform magnetic colonoscopy in vivo, resulting in less pain for patients.
Over 19 million colonoscopies are performed every year in the EU and US, and it is currently the ‘gold standard’ for managing colorectal disease. However, flexible endoscopes are complex devices meaning that it not only requires highly trained personnel, of which there is a current shortage, patients also have to endure a painful procedure as the tissue is stretched by the endoscope pushing through the colon. Furthermore, it also introduces risks such as perforation and anaesthesia-related adverse events.
Now scientists from the University of Leeds, UK, have developed a small semi-autonomous magnetic colonoscopy, using a robot to guide the medical device in the body. They combine the use of robotics, computer vision and advanced control to offer an intuitive and effective endoscopic system.
The robotic magnetic flexible endoscope (MFE) system
The MFE system comprised of a small, capsule-shaped device which was equipped with an endoscopic camera, an internal permanent magnet, functional channels (insufflation and working), and was illuminated by a light-emitting diode (LED). The capsule was tethered to a narrow cable, and once inserted into the body a magnetic robotic arm positioned over the subject guided the endoscope through the body.
The team hypothesised that a sophisticated navigation system that could control the magnetic endoscope by introducing superior levels of intelligence and autonomy could increase the navigational performance, and would be key to driving improved clinical outcomes.
With this in mind the team composed the navigation system into several elementary blocks, organised in three main layers. Each layer provided a set of features characterised by increasing autonomy.
- Direct robot control. This is where the operator had direct control of the robot via a joystick, and no autonomy was available.
- Intelligent teleoperation. The operator controlled where the endoscope should be located in the colon and the system carried out suitable motions of the robot by taking into consideration localisation information and magnetic field interaction.
- Semi-autonomous navigation. The robotic system autonomously navigated the capsule through the lumen of the colon, using computer vision, the user still had discrete control.
In vitro testing
To start the team conducted a benchtop study where 10 untrained users had to make multiple attempts at navigating the MFE in a covered latex colon simulator. They attempted each of the navigation levels 5 times as quickly as possible moving from the rectum to the caecum.
The overall completion rates (navigated in > 20 minutes) for direct robot operation were 58%, intelligent teleoperation were 96% and for semi-autonomous navigation were 100%.
The users found the semi-autonomous navigation method the least demanding and easiest to use.
In vivo testing
The results highlighted the improved ease of use and performance associated with increased MFE autonomy so the team wanted to test the system in vivo, using two female pigs.
Two operators with no prior endoscopic experience, were given 10 minutes to navigate a standard flexible endoscopy as far as possible, then they were given the MFE to navigate using all 3 navigation methods, to determine which was fastest and easiest.
In the first pig, the distance reached using the standard FE was 45 cm but substantial tortuosity in the colon prevented the probe going further. The average completion times were 9 min 4 s for direct robot operation (only 50% completed), 2 min 20 s for intelligent endoscope teleoperation, and 3 min 9 s for semi-autonomous navigation.
In the second pig, the distance reached using the standard FE was 85 cm, this time a blockage prevented any further penetration, the intelligent endoscope teleoperation reached this mark in 8 min 36 s for and 9 min 39 s for semi-autonomous navigation. The lowest level, direct robot operation, was unable to reach the marker.
Regarding user workload, both users found that using the standard FE and direct robot operation were the most physically and mentally demanding and that they found it easier to operate the colonoscope with robotic assistance. Operating the robotic arm manually was challenging as it was not very intuitive due to the nature of interacting magnetic fields and field gradients, this was overcome with the intelligent options.
Conclusions and future applications
The team here have shown that the inherent complexity of navigating magnetic endoscopes with a single external permanent magnet can be overcome by the development of intelligent control strategies. These type of intelligent robotic systems are easier to use than current systems, and therefore require less skill to operate them. It is hoped that MFE could increase the number of providers who can perform colonoscopies, decentralising these procedures away from hospitals and into clinics and health centres, allowing a much improved access for patients and decreasing waiting times.
The team are hoping to trial the system in patients next year or in early 2022 however, this work here represents a milestone into this clinical translation, and is a culmination of 12 years work.
In the future the team will be looking to increase the robot speed to achieve faster motion and further reduce procedure duration. The MFE will have also have to be assessed for acceptable levels of pain associated with the procedure, which has currently not been done. From a wider perspective the MFE could translate to other applications such as gastroscopy and bronchoscopy.
This work represents an increasing range of new AI-driven tools that are designed to help ease the burden of over-stretched health systems across the globe. In the main, they are designed to save time, reduce the demand on specialists, and help the human experts make better clinical decisions. We have recently seen the development of artificial intelligent models to predict early-stage breast cancer, and COVID-19 associated pneumonia. The MFE augments this research, adding another AI technology that can accelerate time to diagnosis.
For more information please see the press release from the University of Leeds
Martin, J. W., B. Scaglioni, J. C. Norton, V. Subramanian, A. Arezzo, K. L. Obstein and P. Valdastri (2020). “Enabling the future of colonoscopy with intelligent and autonomous magnetic manipulation.” Nature Machine Intelligence 2(10): 595-606.
https://doi.org/10.1038/s42256-020-00231-9
