Innovation through exchange

Training robots, exoskeletons, virtual rehabilitation, gamification: in recent years, new technologies and treatment approaches have steadily found their way into neurorehabilitation. They support therapists in treating individuals who, for example, have to relearn motor skills in the aftermath of a stroke. Considerable potential is attributed to these devices and applications: they enable intensive training at an early stage of the rehabilitation process, provide playful motivation and allow for progress to be measured precisely.

As things stand, however, this potential is not being fully realised – a view shared by Martina Spiess and Rudolf Füchslin. The rehabilitation researcher from the Institute of Occupational Therapy and the physicist from the Institute of Applied Mathematics and Physics are leading, together with Anne Deblock-Bellamy from the University of Applied Sciences and Arts Western Switzerland (HES-SO), the project “Data-based Optimization of Outcomes: Successful Technology-supported Therapy (D-BOOST2)”. As part of the project, which is funded by the Swiss National Science Foundation (SNSF), they are developing an AI-supported system that assists therapists in selecting, applying and combining rehabilitation devices with a view to improving patient care. “While the devices relieve therapists of some of their manual work, they usually operate in isolation from other devices and therapeutic options,” says Martina Spiess. In neurological rehabilitation, however, a mix of technologies and therapeutic approaches is always used, which makes clinical decision-making challenging. For each patient, therapists have to continuously decide which devices and interventions should be used, in which combination and with which intensity. “In making these decisions, they have to juggle a great deal of information: individual patient data, expert knowledge of the various devices and their functions, current evidence and general knowledge of physiology and motor learning.”

The AI-based tool being developed as part of the project should simplify this complex decision-making process: It analyses patient data, recognises patterns and provides recommendations whether devices should be used, and if so, which ones and which training parameters should be selected. “AI is intended to support the decision-making process, not replace it,” emphasises Rudolf Füchslin. In order to make AI take on this role in the future, the interdisciplinary project team first had to define the criteria that are actually relevant for clinical decision-making. Following the principle of user-centred design, the researchers have worked closely with several rehabilitation clinics to achieve this. “In the first year, we conducted interviews and a workshop with therapists and defined the criteria through an iterative process,” explains Martina Spiess. The team is currently – once again in close cooperation with the clinics – collecting the data required for these criteria so that the algorithm can be trained.

Developing an AI system that provides tailored treatment recommendations for patients with different diagnoses and therapy goals and that incorporate devices from various manufacturers is a complex undertaking. It is one that necessitates cooperation between various disciplines and collaboration between researchers, clinical practitioners and industry. The project team therefore includes AI and data scientists, researchers with therapeutic backgrounds and experts in complex systems as well as representatives of various device manufacturers and rehabilitation clinics. “A project such as this could not be realised within a single discipline,” says Füchslin. Only through exchanges between different fields is it possible to create something new. “Collaboration makes it possible to combine the knowledge and progress of the individual disciplines.”

The interdisciplinary project also brings together two very different approaches. On the one hand, there is technical research, which, according to Füchslin, too often focuses on averages rather than helping people as individuals. On the other hand, there is the approach used by health sciences, which focuses on specific problems in everyday clinical practices as a starting point and seeks solutions that can be tailored to patient needs – while at the same time dealing with the complex realities faced in the healthcare system. Only by combining these two different approaches can an AI system be developed that offers a genuine benefit in clinical practice by standardising and, at the same time, personalising clinical decision-making – a delicate balancing act. “Without basic research conducted under laboratory conditions, the technologies we need for this project would not exist at all. However, this research has to be complemented by a clinical perspective,” says Martina Spiess.

As essential as interdisciplinary collaboration is for the project, it also presents challenges. “We have to invest a great deal of time in order to ensure mutual understanding,” says Spiess. Collaboration requires the awareness that different disciplines speak different languages. There also needs to be a willingness and an openness to listen to one another. “And you have to remain persistent and keep asking questions when something is unclear.”

At the same time, adds Füchslin, it has to be accepted that it is not possible to understand down to the last detail what the other disciplines are working on during the project. “You have to be prepared to step outside your intellectual comfort zone.” The goal is to understand each other sufficiently to be able to make joint decisions. To foster this understanding, the project leads organise regular workshops lasting several hours in which the team exchanges ideas. “Sometimes, it takes up to two hours before we are all talking about the same thing,” says Spiess. She explains that this process of reaching mutual understanding is time-consuming and sometimes demanding – but unavoidable. “To my knowledge, there are no tools that can shorten this process.”

Despite this, Spiess and Füchslin do not see a risk that the “language” of one discipline might come to dominate or that an imbalance could arise. “The disciplines are not competing, but rather complement one another,” says Füchslin, who adds that responsibilities are also clearly defined: the healthcare professionals in the team are responsible for the requirements AI must meet, while the engineers are responsible for its technical implementation.

The fact that the three projects leads and many members of the team have already been involved in interdisciplinary projects facilitates collaboration. “We can draw on this experience,” says Spiess. From the perspective of Spiess and Füchslin, the conditions at the ZHAW also make it easier to carry out interdisciplinary projects. “The ZHAW fosters a culture that facilitates collaboration between the disciplines, provides the necessary resources and keeps barriers low,” says Füchslin. “There is no strict separation of disciplines here.”