When he was growing up in Budapest, Hungary, Peter Kecskemethy would spend time at one of the city’s largest hospitals where his mother, Dr Edith Karpati, worked as a radiologist. He was intrigued to watch her and other specialists like Dr Éva Ambrózay (seen right, © Akos Stiller) staring intently at a computer monitor, looking for signs of a breast lesion. Fast forward two decades to London where Kecskemethy and his friend Tobias Rijken founded Kheiron Medical, a company with the sole purpose of supporting breast radiologists with machine learning software.
Kecskemethy received a first degree in computer science and software engineering in Budapest before earning his second degree in artificial intelligence (AI) and cybernetics at the University of Reading in England where he worked on artificial neural networks. From there he travelled up to Oxford where he received a doctorate in computational statistics with interdisciplinary training in biophysics, cell biology and genomics. Rijken, from Holland, also has impressive academic credentials, with multiple degrees, including a BSc in mathematics and computer science from Amsterdam and an MSc in computational statistics and machine learning from University College London.
The two young Europeans (seen left, © Akos Stiller) developed their ideas while participating in Entrepreneur First, a start-up accelerator that provides intensive training for company founders. This experience provided them with the skills and confidence to establish Kheiron in March 2016 and during the past 7 years the company has secured regulatory approval for Mammography Intelligent Assistant – or Mia – their AI platform for breast screening.
This past Sunday, The New York Times published an engaging story about Kheiron and its two young entrepreneurs that features ongoing studies at five clinical sites in Hungary. Dr András Vadászy, director at one of the clinical sites, commented that AI software is a “huge breakthrough” and said, “If this process can save one or more lives, it will be well worth it.”
Seen right (© Akos Stiller) are Kecskemethy and his mother, Edith Karpati, who now works as a medical product director at Kheiron. She stated, “It’s so easy to miss tiny lesions,” thus making it difficult for a radiologist to stay focused, and that’s where Mia can make a difference. The application of AI algorithms is now a crowded space, and includes companies from South Korea (Lunit), Holland (ScreenPoint), New Zealand (Volpara), and the USA (iCAD). Mother and son will certainly need to stay on their toes to remain competitive!
We first thought that Computer Aided Detection (CAD) would be a major advance in breast cancer screening, but were, unfortunately, disappointed. I am fairly certain that Artificial Intelligence (AI) will be less disappointing. Computers never forget a case, and are not distracted. I am actually a little surprised that these systems (most if not all use neural networks) are “almost as good” to “a little better” (not so sure of this) than radiologists. I would expect that they should be much better than humans, but this is not yet the case.
The biggest problem with these systems is that the computer cannot tell us why it chose to highlight an area on the image or why it chose to not highlight an area. Apparently, neural networks are true “black boxes”. As with “HAL” in the movie 2001, having an inscrutable computer can be a problem. We can’t give it advice, nor can we understand why it gave us certain advice. Another more practical problem in the U.S. is defining who is legally responsible for a computer’s error? It will be a major advance if computers can be true aids in the detection of breast cancer, but I don’t think we are there yet.