Hello, everyone. Welcome to Deep Learning for Healthcare. This is first lecture, Introduction. My name is Jimeng Sun. I'm the instructor for this course. The outline of deep learning for health care, or short for DL4health is the following. We'll first give a quick overview of this interesting combination of deep learning and healthcare applications, and then trying to motivate this topic by talking about the general healthcare data science, why it's important, and also the healthcare applications with deep learning, and then we'll give a quick overview of all the topics in deep learning, we will cover in this course, and activities in this course. Let's get started. The focus of this course, deep learning for health, as the title suggested, is going to be about two different aspects. On one end, we have all the deep learning or neural network related topics, such as different type of deep learning models and the algorithm for training those models, and then on the other hand, we have healthcare applications. We'll introduce different kinds of healthcare data, and different types of healthcare applications. The course is focused on the intersection of these two. In each chapter, we'll first talk about the method in general, the deep learning method, and then we'll cover the used case and the application of that method in healthcare. After finishing this course successfully, you can do the following. You will be able to understand the healthcare data, understand the important applications in healthcare, and you will also be able to apply and hopefully develop deep learning models, and by combining these two skills, you would be able to apply and hopefully develop healthcare data science apps. So you may wonder, why should I care about healthcare? Well, healthcare is huge,it's huge in terms of cost. For example, the US healthcare cost alone each year is about $3.6 trillion. So that's the spending US healthcare had in 2019, a single year. So how big is that number? That's roughly equal to the market cap of the top two most valuable companies combined. So if you put Apple, Google, Microsoft, Amazon, all together, and just add their market cap together, that's about one year's US healthcare spending, just a single year. So cost is a big problem in healthcare. Well, you may wonder maybe, healthcare is just expensive, you just need to spend this much of money. Well, I can tell you there are a lot of waste in the US healthcare system as well. It's estimated over $935 billion per year were wasted in US healthcare, and it's a combination of unnecessary care, fraud, administrative inefficiencies, healthcare provider errors, preventable conditions, and lack of coordination and so on. How big is this number? 935 billions. Well, that's roughly or is actually greater than 50 years of budget of NASA, and the agencies, and all the satellites and spaceship into the space. That's greater than their 50 years budget. So US healthcare care does has huge amount of waste. Well, you may argue, sure, we have some waste, but at least we have a good quality of care, since we spend so much money and waste so much money, and not really. In fact, there are huge quality issues in US healthcare systems. It's estimated between 200,000 to 400,000 preventable deaths per year, that happen in the US, inside the healthcare institutions. That's about over a 1,000 people died per day because of quality issues related to healthcare. So this are all very depressing, but there are hopes. The hope is data science or in the context of this course, machine learning and deep learning can help lower the costs, improve quality of healthcare systems. There are a lot of opportunities here. For example, if you want to lower cost, you can think about disease detection, especially early detection of disease and prevention, so that you can catch the disease before it become very severe and hence costs a lot of money, such as heart failure onset prediction. If you can detect heart failure earlier, maybe you can stop or slow down the progression of the disease. That's one example. The other example is utilization analysis. Here it's focusing specifically on identifying waste and potentially fraud in all the healthcare spending. On the other hand, we have quality we can improve to get a better care. For inpatients in the hospital care, we can improve things like detecting sepsis. We also can improve care for home monitoring so that we can enable more cares given at home environment. That's a more comfortable and more convenient for most of the patients. There are a lot of opportunities for data science to develop mottos to push towards this goal, lower costs and better quality. To do data science, we need data. Luckily, in healthcare, we have a huge amount of data. If someone talk about big data in a domain such as health care, they usually mention four v's; volume, variety, velocity, and the veracity. That captures four different aspects of what we mean by big data. For example, volume is something we intuitively understand, the size of the dataset. In healthcare, the largest data come from genomic data, sequencing DNA or RNA. That's the huge amount of data even on individual patients. Then there are medical imaging. That's a much more routinely done at all kind of healthcare institutions. That's also very large, high resolution images that need to be analyzed. The volume of healthcare data is definitely large. Then variety. It's a very interesting challenge in healthcare. We have all kinds of data, all kinds of different modality of data. For example, data coming from the electronic health records. That's the focus of this course. We will be dealing with structure data such as diagnosis code, procedure code, medication lab test. We'll be dealing with unstructured text like clinical notes. We'll be dealing with images. We'll be dealing with time series like the waveform monitored at ICU, intensive care unit. There are huge medical knowledge base that integrating all kinds of medical ontology and knowledge base from different types of data together so definitely healthcare have very big variety in term of data. Velocity is another aspect. Data coming in very fast. In healthcare, that's the scenario where sensors are involved. In one aspect, you have the real-time monitoring devices in ICU, so many different type of measurements such as blood pressure, heart rate, electrocardiograms, and temperature. They are all measured in real time and sampled at very high frequency. Those data can be useful for building motto and also using those real time data to make clinical decisions. mHealth, those are devices, on-body sensors, wearables that you can imagine, that you can use at home. That's all kind of different velocity coming in very fast. Veracity is another aspect. Veracity is about uncertainty in the data. In real data, especially in healthcare, there are a lot of noises or missing data. There are errors in the data, there are false alarms. All of those contribute to the noise level or uncertainty in the healthcare data so we have to deal with that as well. So big data in healthcare is very real. In this course, you'll have the opportunity to interact with different aspect of big data in healthcare.
