So potential healthcare applications. Well, there are many papers that has been published using deep learning models for various of healthcare applications for diagnosis, for outcome prediction, for treatment recommendation, for insurance related tasks such as fraud detection. And for pharmaceutical companies task such as drug discovery and public health related tasks. So in all different aspects or different sectors in house care, deep learning can play an important role. So for diagnosis application you can imagine medical imaging analysis is one important classical use case of deep learning. Trying to diagnose for example, in this case diabetes reptinopathy with the image of the eye or you can think of early detection of disease using electronic health records. This paper which will talk about it later in this course trying to do early detection of heart failure using electronic health records with deep learning model. Triaging. When people show up in hospitals, which one should we care first? How do we know the severity of the patient very quickly and decide based on what order we should treat them? Right now this is highly manual task which is time consuming and costly. Potentially we can develop deep learning models to help making that assessment as well. Outcome prediction. We want to know what's going to happen with patients. For example, when we discharge a patient from hospital, whether they will come back in a short period of time, say within 30 days back to hospital, that will be a bad outcome, because that usually mean we didn't treat them well enough, so they have to be readmitted back to hospital. So we can try to build a readmission prediction model to predict that events. And if we are able to accurately predict readmission, maybe we can treat them or prioritize them and follow up with them even after discharge to avoid costlier readmission. Length of stay prediction, when you admit a patient into the hospital, you probably won't have a good estimate of how long this patient will stay there. Patient want to know that, and it's also important for you to staff your care team and also estimate how many bad you will have in near future. So the length of stay prediction is very important for hospital operation. The mortality prediction. Predicting death, that's important because the deaths usually means bad outcome and that's something we want to avoid. And with this and mortality prediction, we're able to identify the higher risk patient, then proportionately maybe we can put more resources to care for those sickest patients. Sepsis prediction which I have briefly mentioned earlier. That's also another bad outcome that usually have mean sometimes happen during inpatient stay at hospital. That's the infection of the blood and you shouldn't need to very bad outcome like death. So if we can predict or detect this type of events earlier and we give the care teams more time to respond. Treatment recommendation. And there it's very important, once you understand what condition the patient has, the next thing is to figure out what treatment you should be given to the patient, such as recommending medications, especially for a complex patient that takes multiple medication together. So it's important to figure out the best combination to give to the patients. And one task that related to that is understand or predicting the drug drug interaction. Because that's usually a bad events when drug drug interact. So if we know, or if we can predict which drug will interact with which other drugs, we can try to avoid that code prescription of that type of combinations. So that's for treatment recommendation, here are some papers that we will talk about later in this course. Insurance application. So here also very important, one is fraud detection. There's all kinds of fraud happening that insurance company trying to detect and avoid, and so that's one big aspect of identifying which medical claims are real, which are not. Then the other important aspect is to estimate the cost, right? So if you want to set up a premium for your care plan, you need to understand how much those patients will cost you for insurance purpose. So an accurate estimation of the cost means a more accurate premium level, so the company can be more competitive and set the right level. So for pharmaceutical use cases, we have drug discovery and development there from very beginning, where you want to figure out what molecules are good candidates for drug development and trials. So that's a molecule property prediction. And that's very early of this entire pipeline for discovering a drug, and then molecule generation is another important related task. That is, if you know some promising molecules but you want to improve some property of this molecule by creating something similar to that but better, and you can potentially use some algorithms to do that. And we'll talk about some of this in the course as well. And clinical trial recruitment, very important practical task for drug development. And there are papers about, how do you find the right doctor to run the trial? Then, how do you find the right patient to trial matchings? So that you can recruit the right patients to the trial. Public health applications. This is definitely a very important topic these days. Epidemiology models, traditionally it's not a machine learning model, it's more physical mechanistic models based on some physical laws, but with huge amount of healthcare data available to the Public Health Department, and we should be able to make better models with data. For example, we can build a predicting model for COVID cases at different locations and different time, using maybe the claims data of the entire country, or we want a predicting hospitalization, right? So that's another important aspects that public health people cares about, then predicting deaths is another important target related to public health. So that's a public health related application. Okay, how can we get there? So here's the roadmap of this course. The first two weeks, we will introduce some preliminary knowledge for this course. Introduction for this week, machine learning basics. Next week, the second week, which will cover supervised learning, unsupervised learning. And week three, we'll talk about healthcare data, all kinds of data in the healthcare industry. And then we'll talk about foundation of deep neural networks, DNN in week four. Then embedding algorithm such as where to work in week five. So this will give you the flavor and also kind of the mechanism for you to think about how to build a deep learning model. After that, we'll introduce some classical deep learning models. Week six, about convolutional neural network, CNN, which has a huge success in analyzing images. Then recurrent neural network for week seven, which has huge success in analyzing sequence data such as time series, languages, and text. And then we'll talk about autoencoder. That's one of the most popular unsupervised method in neural networks. So that's kind of the classical models of deep learning. Then we'll talk about more modern models in deep learning. And week 9, we'll can about attention models. Thus really the foundation for many of the subsequent models. And then we'll talk about graph neural network. Or if you have graph structures, how do you still build a new network model to analyze that type of graph structure such as knowledge graph and our molecule graph. Then we'll talk about memory networks. There are several very powerful state of art neural networks has been introduced, which is in this week. And finally, in week 12 will talk about deep generative model. So that's the roadmap and activity of the course. We have lectures, will talk about deep learning methods and corresponding healthcare applications. Will also provide a slice and give you further ratings. And then we have a homework assignments that are conducive, primarily programming part of this course. So you can get your hands dirty, build some real deep learning models, analyze some healthcare data. There also some written part of the homework, but to kind of assess your understanding of the underlying method. And then we have a big project at the end. So that's a data science project for you to build a report. To study specific housecare related task that interests you. You give a presentations and you also share your code with us or with the community. So resources for this course. We have a textbook being developed and probably will be ready when the course are online, and we have all the video and slides available for you. And we also actively develop some lab sessions for you to kind of self-guided labs for you to do some of this programming materials yourself. Okay, welcome to the course. That's it.
