11: Using Deep Learning in Radiology With Jeroen van Duffelen

15 Minutes With The Doctor: Learn from Healthcare Entrepreneurs and Innovators

In this one, we interview Jeroen, the co-founder of Aidence. Aidence started in 2015 when the company saw the opportunity to apply artificial intelligence technologies to medical image analysis. Learn how Aidence can automatically analyse radiological images, how they raised 2.25 million in investment, and the Jeroen’s thoughts on AI in medicine. What you will learn in this episode: - What is Aidence and how it works - How this impacts the future of radiology - Advice for raising capital in the med-tech world - The future of AI in medicine What is Aidence and how it works Aidence is developing artificial intelligence algorithms that focus on the analysis of CT chest radiological images. The software is capable of detecting lung cancer, because the company used algorithms to ‘teach’ the program to detect different types of nodules. It is a supportive tool for radiologists. The algorithm reads thousands and thousands of images. Currently, 45 000 images with lung nodules have been annotated and categorised. These images have been inputted into the AI algorithm so that it starts to learn the features of a nodule/cancer. It can recognise all the different variations based on characteristics such as shape, intensity of pixels, volume etc. How this impacts the future of radiology Before the radiologist receives the images, it has already been sent to Aidence. If Aidence detected a nodule, it will circle the nodule, annotate it, and provide a report describing the characteristic e.g. type of nodule, size, volume and specific location. Currently, it supports the radiologist, in a screening situation or also normal clinical practice. In the future, for potential specific situations such as assessing a single disorder, the system could replace a radiologist but in most situations the radiologists have the broader medical experience and knowledge which is required. It’s very difficult therefore to train an algorithm to detect a rare disease because there are simply not enough examples to feed into it. Artificial intelligence is currently developed by the use of numerous examples which are required for it to effectively diagnose a condition. An algorithm can currently only do a few specific things, depending on how much data has been inputted for analysis. That is the limitation of AI and why they won’t replace radiologists in the short term. Advice for raising capital in the med-tech world Aidence recently raised €2.25 million in investment, but Jeroen says that raising this sum of money for a pre-market, pre-product, revenue business in Europe is difficult. Some of their competitors in the US and Israel have raised $7.5 – 25 million. They plan to use the money to extend their team and product, apply for CE certification and for distribution to first medical specialists. Their advice for anyone trying to raise money for medical technology is to get as much proof on the tech as possible, before trying to raise money. All investors want to know whether you can prove there is commercial viability. AI or IT in health care is so innovative and new that the business models are evolving. Aidence started co-developing with the market, and it was this approach that convinced investors because in collaborating with the market, there is proof of a clear pathway to commercialisation. Currently, Aidence is being used in several institutions in Holland, and with one radiology company in the US, all of which are in pilot phases. They’re also discussing with several NHS trusts in the UK to start pilots as well. At this point, it’s hard to say how much they will charge for the technology but for one particular disease, the price could range anywhere between €1 to €10 per image analysis. The more disorders they can add to the algorithm, the higher the price will become because the value would significantly increase. The future of AI in medicine In the short term,