Exploring the Horizon — Episode 5
Connecting with China for AI in Radiology — Interview with Catherine Yang, vice-president of Baidu
I interviewed Catherine Yang, corporate vice-president of Baidu since 2017. She’s in charge of the artificial intelligence commercialization. Prior to Baidu she was vice president for Microsoft China and responsible for the intelligence cloud business operations. Before that she was vice president and general manager for GE Healthcare IT for Greater China Region. We spoke about the evolution of the market for artificial intelligence in healthcare in China and globally, about the sharing and usage of medical data and about the value of Chinese algorithms in radiology, locally and across borders. You can either listen to this 30-minute interview (by clicking on the Episode 5 figure above) or read this blog.
Erik: Hi Catherine. I’m very happy that I can receive you for our podcast! Today we’re going to talk about how AI is developing in China, and I would also like to hear your opinion about what the future is of AI and how it will develop further on a global scale.
Catherine Yang: Good afternoon Erik. It’s my great pressure to talk with you and also share opinions and thoughts in those hot topics.
Erik: Thank you very much. Let’s go ahead with the first question — How do you think Catherine, the global market for AI solutions for medicine especially radiology, will develop?
Catherine Yang: I think overall AI in healthcare market will continue a rapid growth. We expected to further grow from the current 2.1 billion in 2018 and reach the point of 36 billion by 2025. In terms of the growth rate, 52% is a very high CAGR during the forecast period. So far we’ve seen a very good growth. Basically I think that a few main drivers are responsible for the AI growth. First of all the huge availability of the big data, and secondly the growing number of cross-industry partnerships and collaborations. In addition, I think that the solid demand to reduce the imbalance between healthcare providers and also patients will continue to be one of the solid drivers augmenting the use of AI in healthcare market.
I think that the solid demand to reduce the imbalance between healthcare providers and patients will continue to be one of the solid drivers augmenting the use of AI in healthcare market.
Erik: What do you think will be the most successful business model for AI? Because I see at this moment everybody’s kind of in doubts about what will be the most successful way to offer AI services in healthcare.
Catherine Yang: Well that’s a very good question. Yes, indeed a lot of radiologists and also healthcare providers and vendors are exploring the business models because in the end we need to see that both the healthcare providers and the technology providers benefit from AI technology.
I think to define a mostly successful business model varies and is based on the different situations, but we’ve seen that there are two business models which are quite common and successfully landing in a lot of hospitals. In the recent past, most medical imaging analysis software companies used a workstation-based type of business model, but once we introduced the cloud-based AI technology, we’ve seen that subscription-based and the fee-per-study models were becoming more and more prevalent. These are forecasted to increase their market share in the coming years. So both the healthcare providers and also the vendors mostly prefer these two models.
Once we introduced the cloud-based AI technology, we’ve seen that subscription-based and the fee-per-study models were becoming more and more prevalent. These are forecasted to increase their market share in the coming years.
For the subscription model there are a couple pricing strategies. The vendor can charge the hospital based on the scan volumes. So it’s a volume-based pricing model with fixed price for pre-agreed scan quota. That would be easier for the hospital to budget for, and then it’s also very good for small startups or companies that they would like to to build up the brand and seed for growth. For the healthcare providers it’s also very beneficial for them to scale up once they see the value from the solutions. That’s the subscription model.
The other model is the fee-per-study model, r.e. Pay-as-you-Go. Customers are charged based on their usage of the software. This is a well established business model, particularly for the neurology image analysis we have seen that this is very common model. With this model, customers can maximize the feasibility and lower the entry cost, and also the risk of their investment in new technologies. It’s also good for the providers willing to build up their brand and seed the market. Often both models are offered — fee per study models and subscription model. Then the customer can start out on fee-for-service study model and switch to a subscription model later on. If we talk about the most successful models, I would say these are the most adopted business models so far.
Erik: What do you think will make AI as widely available as possible? How do you think it’s possible to avoid a discrepancy in the market so that some hospitals or some countries will not be able to use AI and others will?
Catherine Yang: Okay, well, that’s a good question but we have to divide it into two aspects. One is the AI technology that is being accelerated and accepted as a solid solution. The first driver is the big data. We gather a lot of data, both structured and unstructured, not only from the hospitals but also from the personal behavior, data from mobile devices, from healthcare providers, data from the patients’ activities. The huge availability of the data fuels AI growth. Second driver is indeed the development of AI technology. Then we have to mention deep learning technology. The machine learning, also well known as a hierarchical learning, is a typical type of a machine learning that involves algorithms and is based on presentation of learning data. Deep learning has been widely used in the field of medicine, particularly in the radiology area. The radiology department normally is very complex because they are several modalities such computed tomography and also magnetic resonance imaging, producing thousands of images. Deep learning has become an ideal technology to dramatically improve the performance of the computer algorithms in automating the radiology practice. We will talk about that later, but that’s why the deep learning technology has been widely applied in the clinical radiology to detect, quantify and segment the lesions and diagnose and classify the diseases. So machine learning / deep learning is the driver that accelerates the adoption of AI applications in healthcare.
Erik: We can see that in China there’s a lot of interest in these AI developments especially in radiology. What is the main driving factor to use AI for radiology purposes, let’s say in China?
Catherine Yang: Well, I think of course, there are a lot of drivers but fundamentally the medical disparity and pricing issues are the most important ones for both the government and also the healthcare providers to address. AI technology applied in the healthcare industry will be a very good tool to mitigate the existing imbalance in distribution of medical resources. Bring balance between for example, the high-end hospitals and the primary care hospitals, and between the developed cities like Beijing, Shanghai, Guangzhou and the developing countryside, the more rural sites.
AI technology applied in the healthcare industry will be a very good tool to mitigate the existing imbalance in distribution of medical resources.
Erik: How easily are healthcare data currently being shared in China and how is the research with these data organized? Do you think this is done more efficiently in China than the rest of the world? Is it really the ambition of China to become one of the leaders or maybe the leader in the AI market for healthcare?
Catherine Yang: Well, yes the government clearly has a very strong commitment and also ambition developing AI in healthcare. It’s part of the national strategy now. You probably have seen that from the news reports. With regard to the first question on how healthcare data is currently being shared in China, there are quite rigid policies and regulations to mandate the sharing process. In the past, the data standards, regulations and also the definitions varied across regions, which sometimes made the sharing of data extremely difficult, and the challenges were very serious. To address those challenges, a new regulation for sharing healthcare big data was announced recently. The National Health Care and Planning Commission is responsible for developing such a regulation. They will establish an open mechanism for healthcare big data to be shared and promote collaboration between healthcare providers. The National Healthcare Committee is responsible for overall planning, organizing and formulating the national healthcare big data standards and to supervise and guide the acquisitions and access to those data. They also develop a national platform for healthcare big data. These are on their charters. In addition to those regulations, new standards and guidelines were announced.
The Chinese government issued an artificial intelligence plan last year, with a vision of becoming a global leader in AI research by 2030. Healthcare is one of the 4 areas of focus for the nation’s first wave of AI applications (Source: CBInsights)
There are also a few national data sharing initiatives that have been carried out. One very good example is the Chinese Cardiovascular Disease Society in order to promote the cardiovascular data to be shared more safely and efficiently between all the cardiac hospitals. They’ve launched a platform called the Chinese National Cardiovascular Clinical Data Service Platform. The establishment of the data platform aims to actively promote the development of medical informatization in China, and help Chinese cardiovascular doctors and cardiovascular clinics to improve the their diagnosis and clinical research capabilities in the most effective and rapid way, so that affordable health care services can benefit wider population.
Erik: Very interesting, yes.
Catherine Yang: The other national initiative I’d like to mention is also under the State Council and Healthcare Planning Committee. There are five national data centers that have been set up, of which the main purpose is to store regional healthcare big data including data from the EHR, EMR, genomic processing, checkup data and relevant population health data. These regional Healthcare Big Data Centers will help to promote and accelerate the collaboration and sharing of data between hospitals and healthcare institutions. That would definitely improve the clinical research and population health management.
China’s efforts to consolidate medical data into one centralized repository started as early as 2016. The country has issues with messy data and lack of interoperability, similar to the United States. To address this, the Chinese government has opened several regional health data centers with the goal of consolidating data from national insurance claims, birth and death registries, and electronic health records. (Source: CBInsights)
Erik: When China is able to generate all these data and to share them, to promote the development of AI, there will be a lot of new algorithms and new products available in the market. Of course, I know that China is also exporting products to the rest of the world, it’s not only for local purposes or markets. There are some ideas about the usability of such algorithms throughout the world. For example, to cite Eric Topol, he says, “It’s an open question how well algorithms that are trained on Chinese patients and with Chinese scanning equipment will perform when given data from US patients and from US imaging technology”. What do you think about this? Well, maybe that’s the next question, but do you think that algorithms created in China will be usable in other parts of the world?
Catherine Yang: Well, I think it’s a very good question. Because of the quality of data and the high standardization and digitization of the radiological workflow, radiology is prone for AI solutions. This is why AI for medical imaging is becoming the most mature field of AI applications in the Chinese healthcare market. This not like EMR or EHR systems. If you want to launch an EMR or telemedicine solution in a different country, you have to consider the local market regulations and policies for patient treatment and hospital management. The type of workflow driven solutions really varies a lot depending on the needs of the local market, whereas this is not the case for deep learning for AI in medical imaging. Deep learning is very interesting, not only because it brings out greater performance, but also because it doesn’t require a human to identify and to compute the critical features. Instead, during training, deep learning algorithms “learn” discriminatory features that best predict the outcomes. Once we have trained the model and developed the algorithm and we want to apply it to a different market, the data scientist doesn’t need to retrain the model from beginning. The amount of human efforts required to retrain the deep learning systems is less. The DL CNN model is typically designed to perform image classification on very large and diverse datasets. To provide good results, general purpose classifiers are capable of learning an enormous amount of image features. So unless there are huge discriminatory features in radiological operations or patient data, most of the basic DL algorithms should be referenceable and applicable. Moreover, deep learning models can be trained to identify incidental findings, to help radiologists in managing their workload, to enhance the quality of scans, and to reduce ‘retakes’ and thus avoid unnecessary exposure to radiation. In these types of AI solutions, the basic models and algorithms are very adaptable, although minor customization and optimization might be required. When those models are applied in other markets, the local data that fit into this model will of course help us to improve the accuracy and results of the algorithms.
Erik: The quality of the data is important, that’s for sure, but of course the information could be biased because of local differences in the population etc. Nevertheless, if the training data are focusing too much on one type of population then the model would have to be to adapted the other area where the model is being applied, right?
Catherine Yang: That’s right. I agree with you. At present, in China AI imaging solutions are mainly used for screening purposes, focusing on the field of tumor, e.g. lung cancer, and chronic diseases. This is low hanging fruit. This way we have accumulated a huge amount of good quality data, but both hospitals and doctors have compelling needs to use AI solutions to improve their efficiency and clinical research capabilities.
Erik: Do you think artificial intelligence will be beneficial for developing countries or desolated areas or underserved areas?
Catherine Yang: Yeah, definitely, yes. So far, although the AI technology has seeped into the daily lives of people in the developing world, we’ve seen the AI everywhere, in the news, in our homes and in the office and in every industry, healthcare, city planning, retail,…. Pretty much every industry has embraced and adopted AI technologies.
Erik: But do you think it will be useful for healthcare as well?
Catherine Yang: Well, yes, of course, for sure. AI technology will dramatically transform the healthcare industry and be beneficial for both healthcare providers and patients. But also, we’ve seen that there’s a lot of untapped potential in terms of AI usage in the humanitarian areas. The impact could have a multiplier effect in developing countries, especially in the developing countries where the resources are limited.
Erik: You mean limited in doctors and availability etc.?
Catherine Yang: Yes and in healthcare industry. I can give you few examples. By leveraging the power of AI, I think that both non-governmental organizations (NGO’s) and governments can solve their life-threatening problems and improve the society and then the life of the local communities in the developing world. I’ll give you a few examples how we use AI in China in the rural areas. We have a lot of earthquakes in the southwest region Sichuan and Yunnan Province. The provinces have a very high frequency of earthquakes. The governments actually can use the drones and big data analytics and AI technology to detect and predict the earthquakes and also to analyze post-earthquake reconstruction needs, so they can better distribute the resources among the earthquake regions. For example, a few years ago there were a dozen of cities in Sichuan that have been impacted. The humanitarian agencies had to distribute those resources within a very short time, including dispatch army, doctors etc. to save disaster victims. With the use of the big data analytics and also AI technology we can optimize this delivery and shorten the response time.
The other example is very interesting, it’s in agriculture. One of the sectors AI has been widely adopted is in agriculture. Farmers, especially in the rural areas, can use IOT and AI applications to improve the seeding, fertilizing, yielding and crop yields, raise pigs and livestock and boost the profits. Drones have been highly adopted in the agriculture industry.
Erik: I was mainly thinking about using artificial intelligence for areas where there’s insufficient capacity of medical doctors. So that for example images can be analyzed without the presence of a physicist or a medical doctor. That could also be a solution probably.
Catherine Yang: Yes, I totally agree with you. I forgot to mention that one of the key verticals that is highly promoted, supported and also budgeted by the government in China is AI in healthcare for provision of primary care. The doctor ratio in China is far behind the U.S.. We have 1.5 doctors for 1,000 patients. Not to mention the countryside and rural areas. It takes a lot of cost and a long time to develop and train a good doctor; training a specialist doctor takes 12 years. Using telemedicine and AI aid solutions can greatly reduce medical disparity issue and provide timely support to patients in primary care market. A doctor or radiologist can be supported by AI tools that are working as a personal assistant. After the AI-aid imaging solution has conducted a preliminary screening, it provides the suggestion of a diagnosis . The physicians can share the patient’s image report by uploading it through the cloud-based platform, and refer them to senior radiologist and clinical doctor in an upper level hospital for further examination. This will dramatically reduce the doctor disparity issue and also help to get out the result or the report faster.
Erik: How do you see the future of AI? How fast will this technique be adapted on a wider scale in China and also in the rest of the world? Where do you think it will be used or accepted most rapidly?
Catherine Yang: Well, as I mentioned I think AI it’s not just a theory, it has many practical applications. In 2017 market research showed that by 2020, which is not far away, at least 30% of the companies globally will use AI in one of their business fragments. Today I think large corporates leverage AI to optimize their processes and to improve customer services and R&D, and to achieve higher revenue & profits. If you pay attention to what’s happening not only in healthcare, but in energy management, retail, financial services (FSI) and manufacturing, you’ll see that these are the leading verticals that embrace AI. AI will not only become corporate strategy but also become very important for P&L (profit and loss) in the companies, in some large companies. We predict that the AI will continue its rapid growth in the future.
Erik: Is healthcare slower?
Catherine Yang: No, no, I think it will continue to grow. We don’t expect the hype to die down in the next 1 to 3 years. I think we’ll continue the rapid growth and then we’ll get wider applications in healthcare industry.
Erik: Do you agree that healthcare is relatively slow in adapting AI compared to the rest of the market?
Catherine Yang: Yes, I agree with you. I think the rate of adoption may be initially slow in healthcare due to the nature of the industry and the requirements. But healthcare industry is definitely one of the most suitable market for AI adoption, for quick diagnosis, better personalized treatment plan, improved healthcare outcomes for both provider and receiver. Hence we will see a quick ramp up and rapid growth.
Erik: One word I would like to talk about briefly is the “AI Cold War”. That’s a word that more frequently being used nowadays. Aren’t you afraid of an AI Cold War? Do you think markets will become isolated or will this cause an atmosphere of distrust and a mutual suspicion between both sides? What do you think?
Catherine Yang: Yes, this topic is being discussed more frequently nowadays, and a lot of attention is given to this. But I think fundamentally AI technology is for the good of human beings and the entire society. I think the main drivers in China to accelerate AI applications, commercialization and technology development is the huge market needs. We just talk about needs in the healthcare market, which is coming from both healthcare providers and patients. Both Government and companies, private sectors have committed and invested huge amounts of money to accelerate the AI technology development and commercialization. These will definitely accelerate the growth in the Chinese market. From this perspective I’m not afraid that AI will die. As I mentioned, China indeed has a leading advantage in terms of the huge amount of data because of its large population. I’m not sure about the data quality though [laugh].
But I think AI does require collaboration.
Erik: Exactly.
Catherine Yang: Yes, AI can help to break down the silos and barriers, because the collaboration across regions, across the hospitals are indeed one of the drivers that accelerate AI adoption in North America and in Europe. So we look forward to more collaboration for example with European companies, we expect full scope cooperation in all aspects of AI, sharing data, standard development, clinical research, talent development etc.
We look forward to more collaboration with European companies, we expect full scope cooperation in all aspects of AI, sharing data, standard development, clinical research, talent development etc.
Erik: But maybe for facilitating this collaboration there should be a joint effort to develop some trust, and maybe to develop that trust there is a need for some rules and norms that are common, that are shared commonly to promote the development.
Catherine Yang: Yes, of course. I’d like to give further examples. In China there is more than 100+ startups that are actively playing a role in the AI radiology market. Some of them are really young startups and some are big giant players like Alibaba and Tencent. Also image equipment providers (United Imaging) tap into this new market. Like Infervision, Peredoc, Airdoc and Deepwise. These are the leading players who are actively exploring the European market. So I think this is also a signal and sign that collaboration in AI is needed. We are open for any format of collaboration in this area.
E-commerce giant Alibaba launched an AI cloud platform called ET Medical Brain. It offers a suite of services, from AI-enabled diagnostics to intelligent scheduling based on a patients medical needs. Tencent’s biggest strength is that it owns WeChat, which is the most popular social media application in China with 1B users. Around 38,000 medical institutions reportedly had WeChat accounts last year, of which 60% allowed users to register for appointments online. More than 2,000 hospitals accept WeChat payment. WeChat potentially makes it easy for Tencent to collect huge amounts of patient and medical administrative data. This year, Tencent partnered with Babylon Health, a UK-based startup developing a virtual healthcare assistant. WeChat users will have access to Babylon’s app, allowing them to message their symptoms and receive feedback and advice. (Source: CBInsights)
Erik: Maybe the development of some international standards would help in promoting this collaboration on across borders?
Catherine Yang: Yes, of course. I used to be the committee member and Vice Chairman of the China Software Industry Association (CSIA) and the China Hospital Information Management Association (CHIMA). One of major initiatives that I was leading and participating in was the development of data exchange standards and technical specifications, the guideline for e-Health, the big data white paper and the blueprint for digital health in China. We collaborated with key opinion leaders from Europe and the U.S., and invited them to participate as advisor. I think this is a very good example in terms of international standards and in the new area of the AI applications in healthcare.
Erik: Yes, so, actually maybe to conclude we could say that all countries should commit to develop more shared, open databases to promote research and to facilitate the development of AI algorithms. That would be beneficiary for the entire population.
Catherine Yang: That’s right. I have seen similar initiatives in Europe, promoting the algorithms will promote foundation research and AI applications in new industries. It does benefit a lot to a company’s growth and to the entire society.
Erik: Okay, thank you very much Catherine. This was a very interesting interview. We will follow closely what’s happening and how the market will develop but at least you gave us an excellent insight. So, thank you again.
Catherine Yang: Thank you Erik and it’s a great pleasure. I look forward to continuing the discussion with you. Thank you.
Erik: Okay, great. Bye-bye.
