 |
Note: This technology's software is compatible with virtually ANY human language.
Language models can be made or are available for virtually ANY human language.
Concept
Frequently occurring situation:
You are a doctor and you have once again a rather difficult case. After consulting standard medical resources, you wonder if there's a link with the specific population around. You want to quickly find similar cases in the archives of the medical facility where you work, and maybe talk to colleagues who handled them. Thousands of patient files in the facility are available on computer screens, but in not so strictly written formats, containing many sentences in natural English (or any other language). In the best case, the texts are searchable using literal keywords, but the variety of the language used in the files yields a disappointing search result...
So: difficult to find the required information about archived local patient cases, having certain similarities to the actual case you're dealing with.
Wanted: Computer Database in which you can search based on meaning, which should guarantee finding what you're really looking for (if it's there, at least). This database should contain all relevant information, and only that, present in all those patient files, independent of the precise wording and sentence formulation in the original patient files, but only on their meaning. The information should be organized per patient, with an optional link to the original patient file.
Solution:
Technology based on sharp language understanding and artificial intelligence, like Natlanco's: Such a system can indeed read and understand all those original patient documents automatically. It will find the relevant information, and it will put the result into an XML or other database. Medical Personnel can search very efficiently in this database, as it is indeed based on meaning, rather than exact wording.
Result:
Better medical care. Doctors can inform themselves better about patients, medical histories, effects of medication and treatments in the past, specifically for the medical facilities they work in, and the population they treat.
Who can benefit and how?
Who:
Medical facilities, doctors, medical and biomedical researchers, and, of course, patients can benefit. The general public also benefits due to a certain increase of cost efficiency.
Locating specific cases, specific local experts:
Often doctors have a need to search specific patient cases that have occurred in their facilities, as additional information to the existing medical literature. These specific cases have the advantage that the doctors can easily contact the treating doctor and ask further details, so that they optimally can treat their own patient with related problems. This amounts to finding efficiently local experience and local experts for specific cases.
Focusing on the local population and their specific situations and problems:
Also, often the statistical effects of medications and treatments, as well as the occurrence of certain medical problems are slightly different for different populations. This means, it can offer a significant advantage to compare actual current patient situations with those that have occurred in the past in the same medical facilities.
However, often no significant means are available to study the population as well as the much more general populations in the literature. The available documents are in many cases simply the patient records themselves and some rather undetailed statistics, without link to actual patients.
Nature of the Database to produce:
It is clear that it would be of great use, to have a computer system that can examine the records of hundreds or maybe thousands of patients in such limited populations, and that can put every relevant information item present in them into one large medical database, optionally with access restrictions for certain data and user categories. This database should be homogenous, very systematically organised, and should have data that can be linked back to the patient and treating doctor. The data should be based on meaning, not on exact wording. Finally, the database should be easy to search in, with high success rate of finding what is looked for.
The bottom line of the advantages:
-better and faster diagnosing
-better and more targeted treatment
-less cost per patient
-better coordination of doctors and successful treatments
-less medical mistakes, less lawsuits
-higher patient satisfaction
It should be well understood that the systems described here are not in any way a replacement of the medical case literature. But the claim is that they can be a significant extension of the toolset that is available to medical staff.
How is this possible?
The model
The analysis is based on a high-resolution model of the Target Language and Application Domain. This model is a close approximation of the actual language used in the medical records (including the imperfections), but formalized, so that a computer running Natlanco’s software can handle it. A model contains two levels of grammars, three levels of lexica, and a lot of complex details within them. There are also several semantic layers, dealing with the actual understanding. The model has generic and domain specific aspects at almost every layer.
The software technology.
Natlanco's system, a variant of its LingBench technology, reads each sentence or table field. Each word in it is analysed morphologically, each sentence is analysed syntactically, and the result is a set of syntactical tree-structures (parses), with probability information for each possible grammatical analysis. Each of these parses are then also semantically (=regarding meaning) analysed thoroughly, so that independence of the exact wording is established. Finally, the found semantical results are combined with the information in other sentences or fields and the full-document result is written into a hierarchical (e.g. XML) database.
Actually, the system is much richer and complex and has many more layers than described here, but this gives the general picture.
Can the analysis of many patient records be done fast?
Natlanco’s technology is one of the fastest in the market, without giving in on analysis quality. Patents are pending.
Why Natlanco’s technology, rather than that of other companies?
Apart from the processing speed, there are many other arguments:
No other company’s software can understand that deeply the actual meaning. This is due for a large part to separation of software and model. The software sets virtually no limits to semantical resolution. At the same time, modelling a Language is very efficient (regarding development time as well as quality) in the integrated very rich environment of LingBench IDE version 2006.
Total universality toward Languages: models can be made (and have been made) for languages as different as English, Dutch, Chinese, Farsi (just examples).
Open and flexible importation of existing vocabularies with semantical annotation.
Proven although young technology.
Natlanco is still early in the process of building its reference client list, and for this reason, the prices are beyond competition.
Natlanco’s phased contracts system limits your institution’s financial risks drastically, by sharp delivery criteria and linked payments (except upfronts).
The technology offers an ambitious upward grow path. Natlanco’s LingBench technology is compatible not only with existing external technologies, but also with its own full-scale artificial intelligence dialogue system.
Does this technology work today?
Yes, the technology can be demonstrated now. It is however very new, and the first contracts with US based medical companies are in their final delivery stages.
When is this available, what are the formulas?
Any formula is based on a phased contract. The subjects of such contracts are about developing appropriate models for the language and the domain, and customisation and licensing the software. They usually take between 10 months and 24 months. Languages can vary from Hungarian to Chinese. Contracts can be concluded from today. Payments follow as deliveries of the Phases proceed (except small upfront)
Beyond medical record understanding
Grow path toward integration with Natlanco’s Dialogue System, a top-level application of Artificial Intelligence
Natlanco’s Dialogue System shouldn’t be confused with simple question answering system or with dialogue systems based on situation dependent grammar/vocabulary limitations, which might look intelligent at first view but which can understand only rather expected sentences, without really understanding anything else at that given moment.
Natlanco’s Dialogue Technology on the contrary aims for the most ambitious approach.
Basically, Natlanco’s Dialogue System is capable of having a normal, natural conversation with a human.
Again, like in the patient record analysis systems, software and (dialogue) model are separate, so that the underlying dialogue model virtually has no complexity limits.
Of course, usually the model will be focused on a certain domain, although a good model will also deal elegantly with out-of-domain matters.
Applied to the medical field, and specifically to the meaning content database (e.g. in XML) resulting from the system described above, it allows medical personnel to browse in the large database, based on pure content, but this browsing is not browsing in its traditional form, but rather like having a natural conversation with an all-knowing librarian, so that the doctor is guaranteed to find exactly the information he is looking for.
The reason for this, is that the “librarian” tries to keep more or less control of the conversation, in as far as the “doctor side” allows him to (this feels very comfortable for the doctor). In this way, the system gets a sharp idea of exactly what the doctor is looking for, however complex or simple it is.
The system keeps at any moment close contact with the whole of the database content during this process, without the doctor necessarily being aware of it.
The conversation will usually end when the system has found the specific information (if it was present of course) the doctor was after, and the doctor has confirmed it was satisfactory.
Although Natlanco’s dialogue system itself does not include voice (its core is purely textual), it can for many purposes best be integrated with speech recognition and artificial voice, as well as a talking head, which is more important than would be expected: It makes the computer as dialogue partner much more acceptable, and the feedback in the form of head nodding, eyebrow movements etc… contribute seriously to naturality and dialogue efficiency.
At this moment, Natlanco’s Dialogue System Technology is still being further developed (as opposed to the medical extraction Technology which is operational today), especially regarding high-resolution semantics. (Low-resolution semantics are already working). So, for high-resolution semantics, no demo is available yet, but Natlanco already can handle phased contracts including them today, with guaranteed deliveries within a 1 year.
|
