Castellini: Medical technology robotics requires humans and AI

What difference could AI-supported assistance robotics make in everyday therapy?

Prof. Claudio Castellini: AI in assistive robotics is still in development and will likely reach clinics in 10-15 years. Having said that, the main advantage AI potentially provides to assistive robots is to enable a high degree of personalization of therapies. Using both data collected from the patient and the super-large databases it has been modeled upon, it can turn a robot into a companion adapted to the patient’s anatomy, type, and severity of disease, desires, and needs.

How exactly are patients and therapists actively involved in the learning process of the systems?

Castellini: Therapists remain central, monitoring both patient and system. With AI support, one therapist could treat more patients. Patients are involved on two levels: in co-design, to ensure systems are acceptable and effective, and during therapy, by interacting with the robot. This interaction allows data collection and continuous adaptation through human–machine dialogue, leading to better therapies.

What are your three key recommendations for academics that want to design and test AI-based assistance robotics?

Castellini: First, clearly identify from the start what your end-user target population is, and connect to a clinic in which these patients are treated, possibly with a clear track record of cooperation with universities and a strong emphasis on research.

Secondly, instruct your personnel and yourself to deal with human beings and familiarize yourselves with patient studies and their acceptance of assistive robotics. They are more important than the design and testing of devices itself.

And thirdly, the optimal solution is the simplest one that solves your problem. For instance, using a complicated AI method just because it is fashionable or because it raises the number of citations to your papers is nothing but scientific malpractice. In most cases a basic AI method will work just as well, and since it is well known, it will provide much more insight into the hurdles of the problem.

What hurdles still exist in practice and how do you think they can be tackled?

Castellini: The main obstacle is that there still is a lack of connection between engineers and doctors, psychologists, designers, and the “real world”. Although research is not aimed at commercialization of a product, the strong applicative nature of this kind of research cannot be neglected. A device that only works on the bench of your laboratory but then is not effective on or not accepted by patients is, simply put, a failure.

How can hospital managers tell if an assistive robot is truly patient-centered?

Castellini: A successful assistive robot is one that a) is certified safe and can be operated easily by non-engineers, b) sells in good numbers to the clinics, and c) whose effectiveness is proved by a strong successful track record. My favorite example is the Lokomat.

Looking ahead to the next 2-3 years: Which use cases will be ready for clinical operations?

Castellini: I believe we are close to a breakthrough in control of upper-limb prostheses via the usage of ultrasound scanning in place of what is now the gold standard: surface electromyography. Pilots in this area are already appearing in scientific journals, clearly showing that the potential of this technique is immense.

What can visitors take away from your MEDICA presentation to help them take the next steps?