Tuesday, 3 May 2016

Week 4 [02.05-08.05.2016] GAAIN Entity Mapper: An Active-Learning System for Medical Data Mapping

In the world there are constructed lot of expert systems for the medicine. In my article I would like to present one of the state-of-the-art system focused on diagnose one of the worst severe brain diseases. 

http://journal.frontiersin.org/article/10.3389/fninf.2015.00018
http://journal.frontiersin.org/article/10.3389/fninf.2015.00030

Achieving the best possible treatment for the patients require knowledge from all known science domains as artificial intelligence, algorithms, neuroscience, statistics and data analysis. It require discipline and lot of effort to achieve this goal. I’d like to ask you to answer few question after reading of this very difficult article:


  1. Have you faced in your life any medical expert system? Did it impress you?
  2. Is classification presented in this article is based on all important sources of information require to achieve goal?
  3. How do you think this system is secure and produce proper results?
  4. Does this system architecture can be better?
  5. Does active learning system is optimal solution for this kind of systems?

40 comments:

  1. I haven't met earlier with the medical expert system. I think that such a system is safe. the program isn't a doctor. It is the knowledge base - helps a man.
    As shown in the last case of self-learning bot microsoft. Learning systems must be under the supervision of a man.

    ReplyDelete
  2. There is lot of accidents in history of technology that we must remember. However you are right. All software are still on human management. I fear and like to know how this type of software are protected from human errors and you need to be aware that this kind of system are equiped by machine learning algorithms.

    ReplyDelete
  3. I have not used the expert system but when I hear about solutions as GAAIN I think are very important. It's really great that people use computer networks, algorithms and knowledge from many different areas in order to solve such kind of problems. Like Maciej I think that systems are safe as long as people thinking when using them. At this moment it's difficult to me imagine a more interesting solution when it comes to architecture for such solutions. Regards.

    ReplyDelete
    Replies
    1. This comment has been removed by the author.

      Delete
  4. I never had a pleasure to meet with such an expert system personally.
    On the second question, I'm sorry but I do not understand what you mean.
    Question 3 - Why should he be protected and what from ? Please explain.. Probably every architecture of every system can be improved, but I do not see the sense of it, taking into account the purpose for the existence of such system. You can improve a lot of things in this process, not necessarily it’s architecture.
    About question 4 and 5. If you mean to use ML system, then I agree. Using such a self-learning system is always good idea and it improves every solution. Is it optimal? It's hard to say because finding optimal solutions is always a difficult task, perhaps the optimal solution has not been discovered yet. Maybe there is no optimal solutions, who knows ? We always strive for optimal solutions, but all you can do is to keep improving existing ones to make them better.

    ReplyDelete
  5. I think that provided link is no longer valid. Never the less I found it some ware else.
    Any system which help in people treatment is more than welcome.
    Compared to the existing state-of-the-art in schema mapping, the GEM system is better suited to and optimized for biomedical data mapping such as in Alzheimer’s disease research. Their experimental evaluations demonstrate significant mapping accuracy improvements obtained with their approach, particularly by leveraging the detailed information synthesized for data dictionaries. They received high mapping accuracy (90% and above) possible for datasets in this domain.

    ReplyDelete
    Replies
    1. I found on this link very deep analysis of the GAAIN system and it’s true this article focus on Alzheimer’s disease. But I hope don’t wish to scare my audience from this curious article. When I red this article I was continuously wondering about the accuracy and security of system.

      Delete
  6. I wouldn't call the system described in the article (3D-printer visualization of neuron models) an expert system. I have seen some other interesting examples of imaging neurons, like human Connectome Project or Brainbow.
    But answering your questions: I have never seen medical expert system, but I think that it is a good idea. It requires a lot of input data and I'd rather design it as machine learning on big data. Medical facilities should be able to input their cases, diagnoses etc. for the system to improve. System should read data in various formats, like plain text, images (x-rays, MRIs and so on) or audio. Of course learning on the fly is a must.
    Ideal solution should not only diagnose, but also suggest further examinations or test to get more input on particular case.

    ReplyDelete
    Replies
    1. I would like to apologize for wrong link attached to this topic. I accidentally pasted link from other interesting article. I’d like to encourage you to read the proper article but I don’t demand another answer.

      Delete
    2. No worries, I liked the first article more, because of my scientific interest in brain. Recently I'm exploring dentate gyrus and neurogenesis that takes place in this brain region. Thanks for some interesting neuron models.

      Delete
  7. 1. Have you faced in your life any medical expert system? Did it impress you?

    I only read about them, I have never came across working medical expert system.

    2. Is classification presented in this article is based on all important sources of information require to achieve goal?

    I have a problem here. The link you gave points to something completely different and I had to search Google about your topic. I have no idea what article you're referring to, so I am unable to answer this and latter questions.
    Probably I won't be able to answer them anyway, because I have very limited knowledge on the subject, but nonetheless...

    ReplyDelete
    Replies
    1. I am terrible sorry I pasted correct link. I know that this topic is difficult however but I encourage in future to read this proper article or other from this site.

      Delete
  8. Have you faced in your life any medical expert system? Did it impress you?
    Sadly, no. I never had a chance.
    Is classification presented in this article is based on all important sources of information require to achieve goal?
    To be honest, I can only say that 3D printing can help and it was good idea. It's hard to say anything else if you don't understand whole article.
    How do you think this system is secure and produce proper results?
    It's based on some kind of model. Approach like this is always problematic, you are making assumptions and so on. One wrong assumptions and you can throw your model to the nearest thrash bin.
    Does this system architecture can be better?
    As always, but how to make it better is a different question. Without knowledge in this subject it's hard to decide.
    Does active learning system is optimal solution for this kind of systems?
    It's better than nothing. It's slow, needs a lot of examples to learn. However, I prefer doing science this way than not doing at all.

    ReplyDelete
  9. I have never faced this kind of system. I am not sure is 3D printer something what we can call medical expert system...
    But if we are talking about solutions which helps in a diagnose:
    I have never tested this kind of system and I won't meet as a patient (I hope). I think it is very complex and powerful creature. Example of this might be Automated External Defibrillators and this solution I met in health and safety training. It is great solution that might help accident victim better than me. It does not panic, can not forgot about something and in opposite to me: does not faint unless we forget charge this device.

    I can not answer this question in regard to article from link, but I hope that the first answer is in line with your opinion :)

    ReplyDelete
    Replies
    1. I think some read light appear when you read wrong article. Still I encourage to read proper one. :-) Correct link I putted in abstract.

      Delete
  10. Piotr – thank you very much for a very interesting article. This is amazing what modern science can do. Therefore this is sad why modern science can’t provide medical solutions in every kind of problems. Printing neuron reconstructions and artificially generated neuronal morphologies in three-dimensions is fascinating and provide new approaches for understanding the nature and role of these intricate structures. Generally this is my first contact with such a kind of medical software. I reckon such kind of possibilities should be provided for all medical researchers for free. This is the best way to find new solutions for medical problems. In my opinion all countries inside and for example EU inside have their own medical computer system which provide new survey results and new treatments. possibilities to all active doctors and other medical stuff. I reckon this is a shame for humanity that only rich people can treat and can embrace from new medical possibilities. As far as the system from this article is concerned I think this is a great issue but all systems have mistakes and should be updated.

    ReplyDelete
  11. Today′s doctors require decision support aids to help them cope with the management of increasing amounts of medical information (records, research advances, new drugs), make appropriate choices and even to substitute in an expert′s absence. Such aids exist in the form of medical expert systems, which are complex computer programs that emulate clinical reasoning. Expert systems consist of a knowledge base in which doctors expertise is encoded and an “inference engine” which manipulates that knowledge. A number of successful diagnostic, management and combined systems are in use but these are a small fraction of the total available. Preventing wider usage are difficulties in evaluation as well as in response time. Significant improvements in resource management can be obtained by the deployment of medical expert systems, so they are predicted to influence profoundly the future of health care in general practice and hospitals alike.

    ReplyDelete
    Replies
    1. You said about deployment of this system. Yet I’m not aware of the architecture so I can’t tell you is it disturbed or centralized. We all are aware that many developed software are protected by embargo policy and can’t be distributed to many areas in the world. This is obvious but still there are lot of medical specialists who travel the world and serve their services. Right now calculation which policy are secure or fair are rather beyond my current experience.

      Delete
  12. Hey, thanks for sharing with us this scientific paper. I'll try to answer your questions:
    1. No, unfortunately I haven't had the opportunity to face such a system. Have you?
    2. I am not sure whether any important sources of information were omitted for the classification since I am no an expert in this field.
    3. I think that is a question to the authors of the article. :)
    4. I am certain everything that is complex can be improved. At the current state it looks complex to me. ;)
    5. As Kinga mentioned we can't be sure whether it is optimal and under what specification? Speed in treatment, effectiveness?

    ReplyDelete
    Replies
    1. Did you ever go to doctor when you face sickness? ;-) Many physicians are now supporting by expert systems when are trying to diagnose disease.

      Delete
  13. According to first link in article some answers were quite nice :D.
    AD 1. Same as Paweł, I've only read about them and also some YouTube videos were involved in my education at this topic. I think that IBM Watson is this kind of expert system and as far as I remember people are using it as this kind of system.
    AD 2. Same as Michal I ma not an expert in this field so it's very hard for me to answer this question.
    AD 3. I don't have aby idea about it. Maybe in Alzheimer's disease works well but in other cases it's hard to tell. Doctors should evealute accurace of his answers.
    AD 4. I think that everyting that people create can be optimised and tune. So my answer is yes this system can work better.
    AD 5. Maybe yes. If it's results are accurate and his answer speed is good then maybe yes.

    ReplyDelete
    Replies
    1. Yes Watson is today’s icon when we are talking about artificial intelligence achievements. We should remember that there are lot of other systems that are supporting engineers in they common tasks. They are not build for entertainment they are saving human lives and times. But for average people they are treated as a risk for human employment. And finally how many individuals can cure such complex sickness as Alzheimer's. We could hire an army of doctors but even that army won’t have resources to deal with every patient on the world.

      Delete
  14. Have you faced in your life any medical expert system? Did it impress you?

    Yes, I have faced the systems for gene variants prioritization. Their aim is to help for the physician to find genetic variant which may cause the genetic disease. The application compares patient's genetic test results with available databases of known genetics diseases and generates the list of potentially pathogenic genes. They are impressive from one side, because the process of choosing gene is complicated and often not only human genes databases are used (but also mouse for example), but from another they still require improvements, because results from two applications can be completely different or sometimes program can show the gene mutation that does not cause any disease as potentialy harmful.

    Is classification presented in this article is based on all important sources of information require to achieve goal? How do you think this system is secure and produce proper results? Does this system architecture can be better?

    In my opinion every system can be improved and there isn't anything like perfect system, so this one probably also can be better.

    Does active learning system is optimal solution for this kind of systems?

    Active learning system is for sure promising solution, especially when we deal with many patients with different symptoms, which not always could be matched easily in a way that set of symptoms = some sickness.

    ReplyDelete
    Replies
    1. To be honest I was waiting to hear this. The temptation to use artificial intelligence for forecasting data in fast changing environments can lead us to false results. So it is important to periodically measure quality of this systems and react on errors. We hear many stories of the patients that are treated for wrong sickness however this is not the proof to discriminate human or machine.

      Delete
  15. 1. Have you faced in your life any medical expert system? Did it impress you?
    I have never experienced any medical expert system.
    2. Is classification presented in this article is based on all important sources of information require to achieve goal?
    It seems to be right. It will take some to make further improvements.
    3. How do you think this system is secure and produce proper results?
    I am afraid I don't see any drawbacks resulting in the security of the usage of medical expert system. Of course there is a need of an interaction with a real doctor.
    There are many other disadvantages like in the other expert systems like:
    - It takes a lot of effort and cost to prepare a good expert system.
    - It is not as good as having human experts.
    - These systems do not deal very well with ambiguous problems.
    - Unless they get a feedback from the user the systems will not learn from mistakes.
    4. Does this system architecture can be better? Does active learning system is optimal solution for this kind of systems?
    It is quite promising, I think.

    ReplyDelete
    Replies
    1. Thank you for your answer. You said that AI are not as good as human but lets think for while. Computer beats humans in chess when was unthinkable that machine can ever beat human. Today’s the most advanced Go game was win by AI. Even in television Watson won with human in Jeopardy. With time the set of problems where human are better than computers would decrease. There are also another thing that I question. There are sort of algorithms that can learn from set without learning vectors. You can read it further in materials about unsupervised learning.

      Delete
  16. I think that this is the future of medicine. We need to remember that everybody is uniq. Of course there are elements similar or even identical between different people but the combination is often uniq. So why we should use same treatment of everybody ? We should use uniq combination of meds based on uniqnes of each person. Of course we are not talking here about cold or caugh, but for example cancer or other very dangerous illnes. Our technology is moving forward, we have new concepts, ideas, software, we see more and we understand more. We need to combine all this - like they say in the article - and use it to preper better treatment for specific person

    ReplyDelete
    Replies
    1. Classification was I believe invented to better cope of problems that days. Truly specific persons could be placed in the more friendly environments when they can receive proper treatment. I think all beauty of science laid in those uniq attributes. We should gather that data and try to understand instead of algorithmically process.

      Delete
  17. Wow this is an extremely hard topic and a difficult article. I have very limited knowledge on expert systems. Medicine (brain diseases!) and neuroinfortmatics are way above my expertise. I'm not sure if I'm able to contribute in the discussion.

    I have never seen any medical expert system. Basing on the article it seems to be a really important and interesting piece of technology. I guess that once I have a chance to see something like that in action I'll probably be impressed.

    It's hard for me to judge if the classification presented in the article exhausts all the necessary sources. I assume that scientists working on the system have done the best job they could. Also my lack of knowledge on Alzheimer's disease prevents my from assessing if the system can produce proper results. When it comes to active learning I agree with Emilia.

    ReplyDelete
    Replies
    1. In the today’s research battlefield huge impact lays on the sponsors and the worst nightmare is when someone manipulate in the active learning algorithm for specific strategy. I can’t imagine more advanced problem than that.

      Delete

  18. I never had a contact with systems of this type. Of course it is very promising field of science. The question regarding the sources is quite difficult. I am not an expert in this field. I believe that the system is safe, but the final decision should be up to the doctor. Advertising Architecture, of course, could be better - everything in the world can be better.

    ReplyDelete
    Replies
    1. Agree with you that final voice should belong to doctor. However without good cooperation between medical society and software developers can put us all in danger of moral dilemmas front.

      Delete
  19. I've never seen such a system. Maybe I'm too rarely visiting any health care facilities or maybe it is just something reserved only for "chosen" ones. The classification presented is probably being developed all the time, I'm not able to say if all important sources are already being used (probably no). I hope that the system produce proper results. Situation when some serious medical mistaken is done because of error in program sounds like nightmare. Active learning system is good unless it leads to mentioned mistakes.

    ReplyDelete
    Replies
    1. In apocalyptic scenario algorithm could suggest positive results for the bigger set of healthy population and mistakenly suggest unproper medication. I could not agree that this type of critical system do not require improvment.

      Delete

  20. 1 Have you faced in your life any medical expert system? Did it impress you?

    Unfortunatly I did not face any medical expert system. That I think, mayby I had but know I do not remember any.

    2 Is classification presented in this article is based on all important sources of information require to achieve goal?

    I’m not an expert that I can not answer for your question.

    3 How do you think this system is secure and produce proper results?

    That is another difficult question – I have no idea.

    4 Does this system architecture can be better?

    Of course, everythink can be batter – everything what was built by human can be improve in the future.

    5 Does active learning system is optimal solution for this kind of systems?

    Mayby yes, mayby not I am not specialist of that kind difficult thema – sorry.

    ReplyDelete
    Replies
    1. Thank you for effort. There are subjects I also can't present mine point of view.

      Delete
  21. Have you faced in your life any medical expert system? Did it impress you?
    > Not really.

    Is classification presented in this article is based on all important sources of information require to achieve goal?
    > I believe in iterative development. So far so good.

    How do you think this system is secure and produce proper results?
    > Producing proper results and being secure are sounded like two separate issues. I'm not sure if you were asking us whether a non-proper result would generate a non-secure situation ?

    Does this system architecture can be better?
    > I don't have much expertise on this area. So, I wouldn't like to talk pure logic here.

    Does active learning system is optimal solution for this kind of systems?
    > In my opinion, for now it looks like that way.

    ReplyDelete
    Replies
    1. You might be right that I confused word secure with safe. I should refresh my current dictionary from security field. I would like to answer your question but I have problem with understanding word “non-proper”. I'm familiar with world improper. Is it a similar word or is it a negation?

      Delete
  22. 1. Have you faced in your life any medical expert system? Did it impress you?
    No, not yet. However, During my PHD studies I'd like to focus on recommendation systems and medical expert systems are also a subject of the research in that field.

    2. Is classification presented in this article is based on all important sources of information require to achieve goal?
    I'm not sure yet. Sources of information are probably good enough for a proof of concept.

    3. How do you think this system is secure and produce proper results?
    I think that medical expert systems should be used to support medical doctors decision making not to replace ones. Especially when system is not proven yet, users should remember about a common sense and shouldn't rely only on a system

    4. Does this system architecture can be better?
    Of course! Everything can be improved.

    5. Does active learning system is optimal solution for this kind of systems?
    It seems like it does because system should use real life knowledge from doctors and learn from their feedback. Medicine changes a lot and a good system should follow all those changes.

    ReplyDelete
    Replies
    1. When I study field of expert system I realized how we both have difficult study problem. Let us hope we manage to achive this solution.

      Delete