[Music] good morning ladies and gentlemen as said and the chief medical officer for GE healthcare in Europe I'm a radiologist and I keep my professorship at Hamburg University and so I regularly teach to try to stay up-to-date MA but my responsibility is overseeing medical medical affairs medical education everything that has to do with medical 4G healthcare in Europe today it's about the future of artificial intelligence in radiology and let me start by just giving you a couple of numbers the amount of data that is being generated in healthcare is simply mind-blowing in 2010 it took three and a half years for medical data to double only ten years later in 2020 this year it's only 0.2 years this is 73 days if you think about it this is the time from now till Easter so the medical data is doubled that's really incredible there are 5,600 medical journals putting out 800,000 articles a year there is more information in a mammogram then there isn't the telephone book of New York if they're still there still is a physical telephone book and if you think about it a radiologist in a 12-hour shift is looking at 50,000 images only 15 years ago these were like five hundred images so this is really a lot of data and health care professionals clinicians radiologists radiographers have to deal with on the other hand there are a lot of medical errors happening every year 40 million of course not every medical error is fatal but it is estimated that if you if we take the numbers for Europe up to 350,000 patients die every year due to medical errors that happened in the hospital to make this a little bit more tangible 350,000 and people that's that's a city like Venice in Italy or Toulouse in in France gone every year so this is really a big deal on the other hand we have a shortage of healthcare workers again two numbers this year the global shortage is seven million and if we think in 2035 it is estimated there are fourteen million people and missing you know legging in healthcare and this is due to the fact that a lot of staff is retiring and not enough young people you know moving into the profession or leaving for better paid jobs in the industry so this is another big challenge so I don't want to demotivate you but this is like you know the ramifications this is what we are dealing with right and so and I always hear you know we need this and disruptive technology now and and and you know you know what a surgeon needs least is it is it is a technology this which disrupts him from his surgery right I mean the best disruption is is a is an innovation that is non-disruptive that is unfolding it's magic in the background and and not really and you know that it's not not a parent it's it's just inconspicious lee working in the background so having said that we really have to start doing things differently and we also have to stop doing things quite frankly we also have to stop doing things we used to do and I don't know if you know that you share 99% 99.5% of your DNA with a person sitting next to you just look at your neighbor probably hard to believe 99.5% of your DNA is completely identical so that means we differ in only 0.5% of our DNA that's not much right but on the other hand if you do the math this translates into three million base-pairs and i don't think you need to be a professor of genetics to understand that a drug that i use to lower my cholesterol or my blood pressure my high blood pressure has a potential different effect in me than in you in you and in you and why is that because we are so different these 0.5% make the difference this understanding is very important and paved the way from evidence-based medicine toughts personalized medicine we use to diagnose all the patients pretty much in the same way and then the therapy was the same now we are going to a more individualized approach for diagnosing patients and also we have tailored therapies let me just give you one simple example of where we are applying personalized medicine for many years in clinical routine this is a you know the topic is breast cancer you know that probably 15 to 20 percent of all breast cancers are so called Herceptin positive breast cancers the problem with those kind of breast cancers is they are very aggressive and the prognosis is is rather poor on the other hand there is some light at the end of the tunnel because there is a monoclonal antibody it's called trastuzumab Herceptin and which really helps those patients to extend their life the problem is or the challenge that this drug comes with some side effects including cardiac toxicity so you want to be sure that you only give trusted sumup to those patients that are Herceptin too positive because otherwise if you give it to every patient with newly diagnosed breast cancer those patients only get the side effects and there is no effect so that means that in every patient with breast cancer a so-called molecular and analysis is done to really look if this patient is Herceptin positive and of course only in those patients Herceptin is given very easy example of personalized medicine so in personalized medicine we are moving away you know from this generalized approach one size fits all more to a tailored individualized approach in healthcare or you could say we are moving you know from a philosophy where you know every patient is diagnosed the same way to really a tailored therapy not for each individual patient but probably for some cohorts of patients and if you look at personalized medicine there are basically three buckets there is the diagnostic bucket there is the therapeutic bucket and then there is the monitoring part we are dealing with traditional radiology data in vivo data and then we have all these kinds of omics data in vitro data coming from lab from pathology from your wearables from the EMR just one remark regarding wearables I don't know who has a wearable or has a Fitbit or something people who have have a wearable usually don't need it because they are athletic anyway I mean the wearable was invented as an option for athletes and now we have to translate or transform this into a medical device I mean the 80 year old patient with the BMI of 35 sitting on the couch the entire day eating chocolate this patient probably is in need is in need of a wearable or probably it's too late and in this kind of patient but it's it's interesting what is happening with this variable market and when we are talking about the explosion of data and you've heard about it we are talking about exponential growth compared to linear growth and let me just give you a quick example that I think nicely illustrates what exponential growth really means just assume I have a step length of one meter to make it easier so in a linear in linear growth if I walk 30 steps I have walked 30 meters if my step length is 1 meter so in an exponential growth if my starting step length is 1 meter I have walked 26 times around the globe after 30 steps it's really incredible it's mind-blowing and just I want you to keep this in mind and when someone talks about you know exponential growth sometimes they show these graphs they go up and you think wow this is really they really go up but I mean 30 excursions 26 times around the globe so how can we deal with this avalanche of data the poor radiographer the poor radiologists dealing with all these kinds of data so now AI artificial intelligence is coming into the game and before I talk a little bit about AI let me just ask this question and probably it's a little it's little frightening will a I become humans last an intervention last invention because you know from that time on everything that is going to be invented will be Co invented by AI probably if you think about it and as you know we are surrounded by AI in our daily lives who is using AI we are all using AI at least everyone who has a smartphone I guess that's the vast majority of the people here uses AI every day just a couple of example every time we do a google search and click on one of the suggested links we are part of machine learning and Google takes our click as an indication that you know the results proposed were pretty good otherwise who wouldn't have clicked on them and is using and you know all this feedback to make the search and the search results better other examples include Netflix for example every Friday I get an email what to watch based on what I have watched there are other examples if you use Siri uber and also an example from GE health care from our aviation colleagues we are using artificial intelligence for predictive maintenance in jet engines the airline's really love that for every engine that is actually sitting in a plane there is a digital twin a so-called digital twin on our computer systems and as you can imagine a jet engine generates a lot of data in real time and this is sent to our computer systems and then we can really go away from this maintenance after a thousand hours or 2,000 hours of operation more towards a flexible maintenance approach and of course there are lots of cost savings that can be generated and it makes complete sense and if our computers indicate it makes sense to do some maintenance tonight the airlines can avoid technical issues technical failures and with the need to rebook patients and and cancel flights and stuff like that and we have integrated this approach into health care so now let's take a look when we talk about artificial intelligence and imaging analytics in healthcare where can we apply an AI I see three different levels there is the individual level and what I mean by that is that we are implementing AI capabilities right into our scanners in to our CT systems into our M our scanner into our ultrasound scanner then there is the departmental level this is operational AI we use AI to streamline workflows in radiology departments in private practices and then we have the so-called enterprise level and enterprise level means we can use a I and to look at patient flow in entire hospitals or even hospital networks I will come to that later let's start with the individual level as I said we can implement AI right into our machines and I would like to give you an example from x-ray you know that a condition hospital sphere is especially on the ICU is a pneumothorax a collapsed lung and you also know if not diagnosed correctly and in time it can be potentially deadly and if you think about the situation it's 3 o'clock in the morning and the technician is performing an x-ray with a mobile x-ray system on the ICU the radiologist is probably in the emergency room or is reviewing some CT cases so the tech is doing the images the chest x-ray and no one is looking at those images and research has shown it takes up to eight hours till a radiologist actually looks at at this x-ray and what we have now implemented on on a mobile x-ray system is implemented AI capabilities so the technician is doing the x-ray on the ICU and the implemented AI in an alert system with a traffic light you know green yellow red is is really highlighting critical cases so that means the tech can see oh it's very likely that this patient has a collapsed lung and then can send these images to the pec system with high priority so that the radiologist can directly look at those images and what I like about this example and it's not the case whether the AI outperforms the radiologist or the radiologist is still better than the AI it's just a hybrid model you know the radiologists and the AI are working together and the AI is is just highlighting potential critical cases this is a very nice example we have introduced the system over a year ago and this resonates very well and with with our clinician because if you ask them and diagnosing a pneumothorax and it is is is really continues to be a clinical pain point I mean if you look at the image it's not that difficult to diagnose a pneumothorax I mean there there are there are tricky cases where there are several pneumothorax but it's just about highlighting out of those 10 images look at these two first because it's very likely those patients have a pneumothorax so you know the Prince Prince of Wales and if you look at that image well is the Prince of Wales really showing the finger to the reporters probably not you know as a radiologist you always need the second the lateral view and this was just outside Kensington Hospital and you know when and his wife gave birth to their third child and he was just illustrating to the reporters you know now I have three three kids at home why do I show this the best radiologist will miss the diagnosis or will do the wrong diagnosis if wrong images are highlighted so that means we really have to pay attention that the algorithm is validated and is capable of of really highlighting the critical images and not some images you know they look fine and in the end it's the radiologist who is signing you know with you know by signing the report saying that I have really reviewed the images but you know it nowadays you can generate a thousand images in ten seconds and probably a radiologist cannot cannot review all the thousand images so we are having AI to highlight critical cases so this is very important and and that that we know or that we really have to take care that algorithms that we are developing with our partners are really capable of really highlighting the the critical an image series so the second part is the departmental level and as I said we can use a I to make workflows better in private practices in in hospitals in radiology departments and this is just one example from a private practice in Germany in the frankfurt area there is a customer of us and he owns nine or ten imaging centers in the frankfurt area and the waiting time to get an mr for his patient was too long at least he thought it was too long it was six weeks you know if I give this presentation in the UK they would love it it's only six weeks waiting time for him this was unacceptable so the first thing we did we optimized the imaging protocols and this is very important without sacrificing the image quality so we were able to reduce ten times by 16 percent and keep keep the good image quality and then you know together with dr. Alice that's our customer we looked at you know the scheduling system and we looked at the radiology at the risk system and we could actually you know optimize processes here so in the end we could drive down waiting times from six to two weeks and the nice side effect of course if you can scan more patients especially if you're in private practice you can of course generate more revenue this is an example of how we can use AI we call this brilliant radiology imaging insights in radiology departments and the final level where we can use AI is the hospital level or I said the network level and we call this command center this looks like a NASA control room but it's not this is inside a hospital and we are using you know predictive analytics to manage patient flow to manage patient experience in emergency departments and on the ICU this is an example from the UK from Bradford where we recently opened a command center we call this command center we have more than 10 command centers in the u.s. up and running I remember when I was a resident in radiology I did one year of internal medicine and and so I I was in the emergency department and again it's in the middle of the night and you have to find a bed for a patient so we used to call the First Ward and the nurse would tell you sorry we are full you would call the second Ward and it was really tough to find a bed now you have full transparency you can see where are where are available beds where are clean beds and then you can really optimize and you know the usage of of beds and you can translate this there is data from the US that you can add virtual beds just by better using clean or available beds so this is a very interesting concept it's called command center and hospitals and really really like this to really manage patient flow in the hospital so now I would like to come to a very important and point GE is a big company so GE has like three hundred thousand people working for GE GE Healthcare has like more than fifty thousand employee but if you do the math the majority of people is outside G so also the majority of smart people is outside G so we need partnerships we are looking for partnerships worldwide to develop applications and also we need partners to tell us if if the things we are so excited about that we develop are really clinically useful sometimes our engineers are so enthusiastic they think they have developed something great but in the end there there is no need right for it that is why we need partnership partnerships and we need the user experience we need we need the partners worldwide and with regard to the application development we are looking for data partnerships in Europe also in the u.s. of course but more and more also in Europe and there is not one partner for the entire field of application development so we have a partner where we develop this pneumothorax app together this was UCSF we have another partner and and where we are looking for em are of the heart and how to apply ai there so very specifically for certain indications we are looking for partnerships so if you think of the future and the impact of digitization on future jobs so sometimes I'm asking myself so what does it need to thrive and to survive and to have a good career in the future we all know about the IQ you know the intelligence quotient and we also know about EQ emotional intelligence but there is a new term that I would like to introduce to you and this is TQ TQ is the technology quotient meaning how open are you how open are you to embrace new technologies or are you more like the person I've done it v the thirty years like this I will not change of course probably not everything that is going to be developed in the end turns out to be useful but the technology quotient really shows your ability to adapt to to the digitization that is happening around you and if you look at future jobs I mean on the one hand we know that in the coming years every second job will probably be gone due to digitization on the other hand there will be new jobs coming which we probably don't have a clue right now what these jobs will look like but if you think you know of the of the medical minoo area you can think we will need health data analysts we would probably need someone who guides us through this jungle of all the data we will probably have you know prevention specialists who really use data and try to do predictive analytics so there are a lot of probably jobs emerging and I'm sure there will be and I mentioned Google earlier I did a Google search and I typed in a I will replace and there you can see you know the the answers that Google gave me jobs doctors humans lawyers okay and then also radiologists I'm biased I'm a radiologist myself so let's ask the question will a I mean the end of doctors and if you think about what a doctor or especially a radiologist is doing I think it's a complete you know misunderstanding of what radiologists are doing we do much more than just looking at images just think of the exciting field and growing field of interventional radiology where you really work with the patient and within the field of interventional radiology interventional oncology it's the fastest growing field and in radiology or for example radiologists they sit in tumor boards they discuss cases with other colleagues these are all tasks I think and there are not easy you know to be taken over by an AI on the other hand I think it's clear AI can do a lot of great things just think of repetitive tasks or quite frankly boring tasks measuring you know lesions in the lung 30 known Lange metastases if AI can do the job it's great because it frees up some time for the radiologist to look at more sophisticated cases or to actually also talk to the patient so I firmly believe that you know when we look at this that AI is there it's not science fiction it's science fact we have to deal with it but I think and it offers a tremendous opportunity and if we use AI to make a better diagnosis to make a faster diagnosis on the other hand think think think of it you know would you like to sit in a plane without a pilot the autopilot has not replaced the human pilot but has augmented the capabilities of the pilot at almost every Airport for sure in Europe you can automatically take off and land you know and with an autopilot but I mean come on who would like to sit in a plane without without a pilot I like this analogy because the radiologist of course is still is still needed and as I said AI per se will not replace the radiologist but what I also say and firmly believe is that radiologists who do not embrace this technology in the end will we will be replaced by those by those who do so let me summarize artificial intelligence is really here and it is here to stay it will not go away love it or hate it it will not go away I think it it really can help us to see more to diagnose disease faster with a higher accuracy and it will really help to re-establish a human connection between the patient and the doctor so it will help really to humanize to humanize and rate rate radiology so in the end my suggestion would be to responsibly embrace AI and not fear it and with that I'd like to thank you very much for your attention [Applause]