Providing artificial respiration after an accident or during anaesthesia is not entirely easy, especially for healthcare professionals who have not had much practice. In addition to knowledge, it requires a great deal of manual dexterity, which can only be acquired with experience and appropriate training. The Zurich-based start-up aiEndoscopic is developing artificial intelligence-based assistive software that supports medical professionals support when they are inserting tubes for artificial respiration.
Picture: aiEndoscopic
Artificial respiration is a major challenge, especially for medical staff with little practice. Studies show that it takes an average of 200 intubations for the learning curve to flatten out. If breathing stops for longer than a few minutes, for instance after an accident or during a general anaesthetic, permanent damage can occur. The consequences can be very costly. This is where aiEndoscopic comes in.
The start-up company is developing assistive software for intubation devices. Philippe Ganz, CEO and co-founder of aiEndoscopic, compares intubation with parking a car: Old vehicles had to be parked by the driver without assistance systems. Later, reversing cameras were added. For a few years now, there have also been park assist systems capable of automatic parking. The situation is similar with laryngoscopes, the devices used when inserting tubes for artificial respiration. While these devices used to be entirely manual, camera-equipped devices have come onto the market in recent years, making it possible to see inside the throat.
The Zurich-based start-up aiEndoscopic is going one step further, by developing a kind of park assist system to help position artificial respiration tubes. The software being developed by aiEndoscopic utilises so-called ‘computer vision’ methods. Computer vision is the branch of artificial intelligence that is used to teach computers to see meaningfully, i.e. to identify relationships in image data. Video information captured by the laryngoscope camera provides input. The aiEndoscopic software processes this visual information in real time on a small computer in the device itself. On the screen, it shows the person inserting the tube not only the video information, but also whether the laryngoscope is optimally positioned and whether the tube has been inserted in the trachea and not the oesophagus.
The story of aiEndoscopic began in Zurich’s university quarter, as a spin-off from ETH Zurich, the University of Zurich and University Hospital Zurich (USZ). The original idea, which came from USZ, was to completely automate intubation. Philippe Ganz says that it became clear during the development process that it was difficult to obtain certification for an intubation robot of this type, as the more functions medical devices include, the harder it is to obtain approval for them.
For one thing, it must be proven that each sub-function performs its task and is safe when used correctly. Demonstrating the safety of machine-learning systems is just as challenging as developing them. For instance, it is necessary to show which data the artificial intelligence was trained on and to prove that the data set is representative. Assistance systems are in a lower risk category than intubation robots, which is why it makes sense to take one step at a time and proceed iteratively during development. For this reason, like in the development of automated vehicles, an assist system was built that initially only instructed the person performing intubation on how to insert the laryngoscope.
For the device to give the correct instructions, the neural network has to be trained. At aiEndoscopic, this is done by labelling video recordings accordingly. Experienced specialists watch videos of intubation procedures, and identify what happened and how. This enables the software to learn the various relationships between image and instruction. Collecting this data and demonstrating the usefulness of the device requires clinical studies. Such studies are usually carried out in cooperation with hospitals. As there is little innovation in airway management, many studies are similar, so there is great interest in clinical trials of a new device.
Medical technology is subject to strict regulations. The EU’s and USA’s rules are particularly important, whereby the USA is the more interesting market for aiEndoscopic. On one hand, this is because the US regulations are more clearly defined and authorisation can be obtained more quickly there than in the EU. While a single certification is sufficient in the USA, EU member states have often formulated their own additional requirements that differ from country to country and have to be fulfilled in addition to the pan-European requirements. This makes authorisation procedures in Europe extremely lengthy and difficult for SMEs. On the other hand, the market for respiratory management systems in the USA is much larger than the European market. For manufacturers, this further adds to the appeal of serving the American market first.
The USA has a simplified procedure for products that constitute an extension or regeneration of an already certified older product. The software from aiEndoscopic is one such regeneration that could benefit from this simplified procedure. Thus, it would no longer be necessary to have the entire device validated and certified, but just the new function. As long as this function is limited to supporting the professional performing intubation, such authorisation appears to be possible.
