DeepLearning.AI Learner Community Event ft. Ignacio Martinez

Deep Learning AI online series for AI for Medicine Learning Community. I'm very honored and excited to be your guest speaker today. I'll be speaking for the first 10 or 15 minutes and then I'll be answering your question on Slack for the rest of the time. And don't forget to post comment, question and Upbot question on Slack. Well, I would like to do my presentation first off, sharing my screen. Well, I hope that everybody see this. Well, first of all, I would like to say that I am Ignacio Martinez Soriano. I am computer science graduate. My specialization is in artificial intelligence. I am living now in Lorca Murcia. We are in Europe, in Spain and here in Spain, in the southeast. We are in the region of Murcia and in the little town that the name is Lorca, why I'm here. So I invite you to follow my journey, to break into AI. This is my journey, this is my personal journey about how I find the exciting world of AI. I was born in 1976. I am a little old, but I'm here When I was in the high school, I have watched a film that the name is War Games. I think that more of the people that are listening to me know this film is about a teenager hacker. And then when I saw that film, I inspired me to be when I grow up computer science. So I discovered the hacker style that for me the way of life of hacker is people that like the challenge. People who is. Who can. Who can find, who can resolve the problem is making solver. When I finished my computer science graduate in University of Granada, I have learned some basic knowledge. Programmation, algebra, calculus, statistics. The basic of artificial intelligence. And in artificial intelligence we learned how we can represent the knowledge with the all kind of logic. How the inference engineer can choose the rules to using that representation of the knowledge solve some kind of problem. So we learned to be a problem solver. In 1995, I get my job in the hospital and I was the information system management for all the hospital. So my role and my specific tasks were was the IT administration database to administration network. We work with a protocol that is ITIL and we use Scrum. At last of my time there working like it, I implemented the Scrum Kanban to detect to improve the task. They are in the work of the process. And during this time I was more administrator on management project. But I have no any kind of programmation. I spent too many time without take nothing of code. Only that is managed reposition. In 2012, I tried to finish my PhD that I was doing with artificial intelligence and the main areas was natural language processing and clinical terminology. At last I couldn't finish my thesis. But this was a step to begin again to be excited with the artificial intelligence. When I was. When I was. In 2014 I changed the role in the same hospital and I get out of the IT department and I begin to be data analytic. My main job was to work with business intelligence, data warehouse and managing record for decision making. That is the classic situation when you need to work with the management of the hospital and take decision with the data. But really my breaking with the AI was in 216 because. I found the machine learning course at Coursera with Andrew Neng was a big break, a big step for me because. I could translate the academic learning from the university to the practical situation with implementation, with real implementation. In this course that is very. The. The baseline of machine learning is the program code is the language is Octave. But so I was thinking what language I would like to learn and what is the best language to learn to do for my career like a data scientist. When I was thinking I choose between Python and R. So I know a little of R. But the final truth was Python. In 2017 we changed again the role of the hospital and I began to be a really data scientist. But because I add in AI task developing predictive algorithms in the process of the sun situation inside the hospital. My relation in the hospital was not only work every day to get the data to get the clinical reports. I have a research line, I have a part that we need to investigate to research. In this time I made my first research paper that was a clinical nerve clinical named entity recognition. And in 2018, in 2018 I begin to collaborate and to work in the same time like with a hospital and in the same time with a novel startup that this business was about the implementation of prod that was developed with artificial intelligence. But the target was the medicine. I was working there like data scientists. My specialization was the natural language processing. In the same time I was learning always my life is always learning. I began to. I did the deep learning specialization from Coursera in 2018. In 19, sorry, I go with the other specialization of TensorFlow in practice with Coursera my research way, my research career I was doing different international congress. One of the last of the international congress was in cbmes that is a computed basic medical system. And I did a paper that the name is snomed to back that it's a new solution to integrate the word embeddings, the technical embedding with A clinical terminology system that is SNOMED. Then in 2020 I began to collaborate with like Alpha tested consultant in deep learning A and begin to be a associated professor in computer science and system in the University of Murcia. And so I still learning with other specialist course that is about how can I learn a Bert with the mind. Then now I hear my process in all this journey is always learning and doesn't matter the old that you are but always always learning what I working right now in the hospital? Well, I have two line of work daily. The the classical hospital work is using the business intelligence. I need to use a microstrategy platform and I have integrate a platform with a data data warehouse with a Pentaho that is a proud open source. And I made dashboard and clinical reports to the decision making for the management of the hospital. All kind of data exploration is from the hospital information system. I produce clinical management report for clinical service and economic and financial management. From the other way I have the research hospital work that I made some predictive algorithms. For example, one of them was an algorithm to identify the collapse of patients in emergency service. That's algorithm predictive. I have made with Microsoft with Azure machine learning platform that is very interesting to do that and produce an API to get the data. But my specialization and my main target is the natural language processing. So I developed some clinical nerve to integrate the clinical concepts and to tag the clinical concept from the free text. I developed the last year in the International Congress that I said the approach that the name is snome tube. That is a situation to mix the word embedding technology clinical terminology and standard clinical terminology that is Snomed ct. I work with the analysis of discharge reports to identify clinical concepts. All of this is in free text. When you One of the question. One of the principal main somebody know that I'm working in the hospital is that everybody wants data like you know the principal. The principal way to to get knowledge is first the data. We never can forget that to get the knowledge we need data then transfer to information and then will transfer to knowledge. But first is the data what data I have or I can use. I use the data from the hospital information system but all the data is approved from the ethical Research Hospital commission. So the only way to do some application of AI is about a research project. I need to propose a research project and then the commission if pass this project I can make and can develop this project the situation in the hospital. We have a hospital information system that we integrate all the electronic health record in test and we have different other different information system from laboratory, from radiology that integrate show the information and the data of the electronic record. We have image with the radiology image from epax but we have here, for example, here in my hospital we have a problem. A problem that for the situation of the structure and the kind of radiology information system I can't use that radiology so I'm losing the opportunity to to apply deep learning algorithms and investigate in that field. For that case, I prefer to improve my natural language process techniques and try to extract the information and the knowledge in that field. What is the most challenging thing in my work? Well, I don't know if you know that the role or the kind of data scientist in Spanish public hospital and in general in the government administration there is no specific role for data scientists don't exist. It's a pity. But now we have a computer science role. We have statistic roles, but we have data center that integrate all the mathematics, statistics, computer science in the same role. So I tried to create the role for the hospital. One of my challenge was to convince to demonstrate to prove that we can do some predictive algorithms that improve the natural work in the hospital. So my challenge is to have the recognition from the management to work in this role. And to do that I developed the predictive classification algorithm to predict if will be a collapse in the emergency hospital service. With anticipation of the hour we can see if the demand of the patent is so much Our resources are limited. We can show an alarm and warn of the collapse. I made that in Azure machine learning. So for the management of the hospital was a good idea, a good option to take the recognition from the management to be a data scientist. Other important challenge was in the last year in the International Congress my paper is NoMetrobec. It's a new approach to mix the word embedding and the clinical terminology concept. I tried to put a list of research course that I have made what is the interesting to be But I think that is very short. I would like to improve and give more links of my own experience in the Slack channel. I will try to put here the the principal blocks that I used to to read because one of the things more important in this field is always always read, always read and always learn. So it's very necessary to be a list of too many blocks always very important blocks that there is in the in the web and it's very interesting to be always learning the last news One of the things that when you want to be a Data scientist and data center applied to the medicine is to have a good repository at GitHub. Here I can share my repositories that I have too many repos and we have all kind algorithms from machine learning, deep learning very different repos that you can use and you can share and improve. Other classical web form for the people that have research field is the web research gate. And of course one of the classical information that you need to have with your CV is in LinkedIn. Here you need to. It's very important to have these this perfect this profile because you can now you can share your knowledge there in that way. Here I would like to do. I would like to show a demo. But like at home I have only some presentation of my research field and the other international congress that I was made. So I would like to show one. I'm going to try to explain how can develop a research project inside the hospital. I'm going to choose the other one here. This was from 2017. I developed a semantic tag medical concept using word to read representation. Like you know I here from Murcia and this is my hospital. And the goal of this is use the electronic health record like freetest with a codification process normalize the electronic health record to do this. Normally usually we have an expert human and some automatic tools to do this process. The classical tool are named entity recognition and with a semantic tag. The classical semantic tag that there are in the in this kind of research is a semantic tag that is a process to associate a relevant from a ontology with some document. And there are some tool from the web that you can find and it's very interesting to to prove that is matlib, Metamap, Knowledge Map and ctag. All of this we propose a new approach that we use the word embedding model word 2 vet to represent the word of the text by vectors and identify the semantic relation between there. So one word. When you can represent a word like a vector and then we can use the similarity distance to identify that word and we can get the knowledge of the vector. Then we can map with the standard clinical terminology. This is the main of my research. For the people that want to know word embedding. There is a Spanish proverb that is very useful to identify and to define what is word embedding that is tell me who are your friends and I tell you who you are. So this is the basic idea of word embedding is to identify world from a word. What is the word that is around this word in a text. And then I can get the knowledge of this. To do this we use the technical word two veg. That's from Micholov and when he was in Google. This is the structure of the word to be that like it is very technical. I'm going to pass, I'm going to get here. I would like to show you that the principles. The principal function that I use to identify how is the meaning of the word is using similar function similarity that is the causing distance from word vectors. To use the clinical terminology standard I used Snomed CT for the people that work in the hospital. I'm in the research medicine way. Everybody knows the standard that is stomach ct. This is like an ontology that represents the clinical concept with a unique identify. We can get the sacred clinical concept that is in inner text. This is the principal structure of the Snomecity component that we have a description, a concept and a relation. We use this to integrate with the system word to work. At last we did the implementation. You have here some example. Sorry for the sample because it's in Spanish. But what I want to do this is that we can identify the abbreviation of the clinical text and in a free text of a clinical report with the process we can identify the abbreviation and we can map a correct concept of snobaesthetic. This is like for example to identify a fracture of a bone we have only put FX and then we can map with the correct nano fracture for example here to identify HTA and hypertensive. So this is a other one is for example DECA is the appropriation of. Right. And here we have the. The situation that deja is the second place. The second part with the applying other algorithms we can choose the correct. The current number. At the last we have a. A big vector map with the word of the test and we can see the relation between all the. All the. All the words. And then when you have this work with the correct code of the clinical of the Snomed so you can get more information, visual information from the free text of the clinical text. Well, nothing like you can see this. I'm going to share this in the Slack channel. And then when you want I'm going to begin to the session of question and answer. I'm going to stop the AI. Well, I'm going to say if there is some question here, Some partner. I'm going to begin. Daniel Verdu say for how is collaboration between private company public healthcare system in Spain our main improvement in Spanish healthcare driving by doctor and university research of prepaid company or Both. Well, this is something complicated because to be a collaboration between the public company, sorry, private companies and public health care we need to do research project. When the research project is approved with the clinic, with the ethics commission, then we can develop the project and then we can collaborate between the private company and the public health care. So there is the problem like always with the data, with the privacy of the data. But so now the only situation that we can do is develop a research project and then we can have a collaboration between this company. Here in Spanish There is not too many governments, local or regional government that have a platform to share and to invite the other people and the other companies to develop research project. So we need to be in contact with the university and the university need to know the people in the. In the private company and then is necessary to develop a research project and then we can collaborate. Samyak SA is a solely able to replace the human need on some part of the medical diagnosis or in to help the doctor. How much has the implementation been? Well, This develop that I show is about a tool to help the medical specialist is a tool to. To help the codification specialist, the medical people to get the correct code of the clinical concept. This is only in a research way. I have a test to do in real cases, but in my case only I have for the research and I have in production to all the hospital. I hope that in the next year. Sorry, in the next month, more or less I have a product with the. Oh, okay, sorry. I'm going to see this. I think that this is a little slower. Maybe now it's better. So I Now it's only to reset for the research way. For example, Rahul Kumar I'm working on identification and detection of percentage of burn on patient. My mother doesn't behave well with respect to patient with dark complexion. How do we address this kind of issue? Well, I don't know if you have a. I don't know if you have the real. Exactly. I need more information about your project. We can chat, we can discuss that in the channel. In the channel of slack when you want. I'm going to see. For example, have you heard about all the initiative of using AI to make diabetic people life better? I already know about detecting diabetic retinopathy and it is difficult But I've seen more unused in US and Europe and particularly in Spain doing some research about beta Cell therapy type 1. This is A. This is a good research line. But here in Spain, where we. For example, in the renal. In the renal and health care system. We have some research line about diabetes, but not knowing this using the AI I would like to do one line about that but now there is no reset light explicitly in AI projects. Sorry. Arturo Polanco said that what is the meaning of pacs? PAX is the repository of the image is a big database databases that you have all the radiology pictures and then you can choose one one to one that what picture of the of the back of the system that we have stored is a big database store of radiology picture. There is other for example Nitin I would like to know that use of instrument into Word 2 veg is the implementation used in query generation and early detection of disease. Really the implementation is how the tool comes extract from a free text clinical report the concept, the tag of the concept and identify what is the clinical concept. Within Snomed CT. We have made a prototype that is like a shirt like Google Search that you can introduce your concept and then look for into the clinical text clinical reports to match the correct concept in that clinical report. But with the semantic way. Going to do the last question and if you want we can go on in the Slack channel answer your question and discuss what do you want here? Erwin for Dr. Exploring AI in Med. Is there a general pattern of course layout? You can suggest that we focus on the data science on the computer science behind AI. Thank you. Well, for doctors exactly is a little. Not a little complicated. But is it necessary that no more technical approach because when you develop some algorithm in AI in deep learning or in machine learning, you need to know for example Python, you need to know different libraries. You need to know something that is necessary to learn if you want to understand what are you doing. So one opportunity is perfectly the deep learning AI machine for medicine. I think that there is other one another part now I don't remember, but I think that I have a list of resources in my presentation. I going to put other one in the future and you can choose what kind of course you can choose to improve your artificial and intelligent techniques. Well, nice to nice to everybody for this keynote for spend your time in my presentation. I happy to to enjoy the this situation.