Our 9th daily Comments Competition winner is Falafulu Fisi, for his comment on our post Web 2.0 Meets Medicine. Falafulu told us that the "current state of the art in medicine 2.0 of today is the automated online CDSS (Clinical Decision Support Systems)", which he says is starting to do diagnosis via the Web. Congratulations Falafulu, you've won a $30 Amazon voucher, courtesy of our competition sponsors AdaptiveBlue and their Amazon WishList Widget. Here is Falafulu's full comment:
"Medicine 2.0 ?
The current state of the art in medicine 2.0 of today is the automated online CDSS (Clinical Decision Support Systems ). A doctor in a physician in rural clinic could upload medical data , such as lab tests, patients personal infos (age, ethnicity, alcohol consumption, smoker, etc,...) and so forth. These medical data are queried the CDSS from a central server somewhere to give diagnosis (specific cases only and not general) of the situation based on the patient's data which has just been sent thru.
Another type of unstructured medical data is if a suburban clinic with a facility for MRI medical imaging, EEG & ECG readers, etc,... could just take scan the patient and upload those images or EEG/ECG to query the CDSS to give a diagnosis of the current image/EEG/ECG that has just been sent thru based on similar images/EEG/ECG signals that have been stored & index by the CDSS.
CDSS is not new, it has been adopted in clinics and major hospitals over the last 30 years or so, but this time, it is moving into the internet, ie, automated diagnosis anywhere at anytime. CDSS is frequently covered in certain issues of the Journal of Artificial Intelligence in Medicine.
I am currently developing a small CDSS for automated EEG/ECG diagnosis, using DSP (digital signal processing) & machine learning algorithms (this is my hobby). This app is to be deployed at Auckland Hospital's Liver Clinic Unit, for internal use only and not web-enabled. If the staffs at the clinic find it useful, then perhaps I will look to further develop it for commercial use, where I will include MRI image diagnosis (image retrieval classification) system and other specific diseases. There are tons of publications in these areas.
I agree that medicine-2.0 is on the horizon where it will be widely adopted."
Comments
Subscribe to comments for this post OR Subscribe to comments for all Read/WriteWeb posts
Does Falufulu Fisi have a blog? I see his wise and informed comments and commentary everywhere but here and would like to!
Posted by: Lucindigo | February 19, 2008 8:03 PMLucindo, no I don't have a blog. English is my second language (I am a Pacific Islander who is a resident of New Zealand) and to be a writer as in a blog, you need to be very good in your writing (grammar, etc...), which is something that I don't have. I just like commenting on other people's blog, since most readers won't come across my comments, since they only read the main post and not the comment section.
Posted by: Falafulu Fisi | February 20, 2008 12:25 AMWeb-based diagnosis is already here to a degree.
Posted by: David Rothman | February 20, 2008 7:13 AMYou might be interested in checking out American Well, a new stat up based out of Boston.
Posted by: zach | February 20, 2008 7:27 AMHello David Rothman,
Yes, freeMD looks similar to automated CDSS, however CDSS is an Expert System (ES) which has a (medical) Knowledge Base (KB) to store facts about a specific medical domain and a Inference Engine (IE), to infer conclusions from the KB as its (CDSS) working horse.
I am not sure if freeMD uses both a KB and an IE or it is still using the traditional procedural programming (specifying the steps the program must take to reach the desired goal), while an Expert System such as CDSS is Declarative Programming (target to achieve the goal rather than how it gets there, ie, it is goal oriented ). This is closer to how human thinks in deriving conclusions based on some given facts about the external world. The other advantage of declarative, is that the IE engine can both reason forward as well as reasoning backward (Forward Chaining and Backward Chaining). You can't do Forward Chaining and Back Chaining reasoning using the traditional procedural programming. It is inefficient and error prone.
Mycin was one of the earliest CDSS that was developed at Stanford in the 1970s. It was reported that Mycin was bench-marked against human doctors which it outperformed the humans, in terms of correct diagnosis and treatment recommendations. Mycin was never deployed as a commercial tool, since it wasn't clear who was responsible, if Mycin made an fatal error in its diagnosis, then who do you sue? After Mycin, many CDSS tools in specific medical domains have sprung up over the years for deployment in clinical use. Today, these systems are advanced, but they also get better and better (in terms of performance accuracy) as time progresses.
Just in case you're looking for some tools to further your freeMD development, check out JESS (Java Expert System Shell), which is an API that I use. You can develop any expert system with JESS. JESS rule-engine API is a commercial tool but you can download it to play with. JESS also has a fuzzy-logic extension , FuzzyJ. There is an open source Java Rule Engine called Drools, which I've never used it. Unlike JESS, Drools doesn't have a fuzzy-logic extension. Fuzzy Expert System is more robust than traditional Expert System that is based on boolean logic (either true or false, yes or no, 1 or 0, etc...). I noted that freeMD, the questionnaires must be answered in boolean (yes or no). The problem with boolean logic is this:
Take for example, the Diarrhea rule example in freeMD:
if age is 40
and gender is male
and abdominal_pain is yes
then patient_has_diarrhea
Here is the problem, what if the patient is 39 years or 41 years? What if the abdominal pain occurred once only in the last 10 years? A human would recognize that 39 is closeness to 40 or 41, but in a boolean world, they're not. This is the shortfall of expert system that why fuzzy expert system is more robust than boolean counterpart. The example above can be restated in fuzzy terms as the following:
if age not_too_old
and gender is male
and abdominal_pain is occurred_frequently
then patient_has_diarrhea
The variables not_too_old and occurred_frequently can be defined as fuzzy-sets (set of numbers that ranges from 0 to 1) by the user (the one who designed the expert system). In this way, human ambiguities is captured by the systems, where those ambiguities is completely missed by the traditional (boolean or true-false) expert system.
Posted by: Falafulu Fisi | February 20, 2008 12:14 PMZach,
The American Well doesn't seem to be an automated medical decision support systems. Refer to my previous message above.
Posted by: Falafulu Fisi | February 20, 2008 12:19 PMHi Falafu-
Interesting...but I didn't claim that FreeMD is an automated CDSS.
Posted by: David Rothman | February 21, 2008 7:34 PM