mri

Far from the roots of the first Genetic Fuzzy Tree that focused on Unmanned Combat Aerial Vehicle control, Psibernetix’s methodologies have countless applications in the biomedical field. One such application area is when vast amounts of patient information is known, but what to do with all of the data is not. An intelligent system can be designed to expertly analyze all types of patient information to aid in their medical care.

We have been selected to team with University of Cincinnati’s Professor David Fleck, College of Medicine, and Professor Kelly Cohen, College of Engineering & Applied Sciences, to develop an Intelligent System for neuroscience applications.  Our first bioinformatics system is LITHIA, the LITHium Intelligent Agent (below).  Additionally, the joint UC-Psibernetix project titled “Prediction of Symptom-free Traumatic Brain Injury using Standard Clinical MRI” was one of the two 2015 UC Sports Health Innovation Award winners.

Any type of information regarding a patient, even their genetics, can be handled by this our methodologies. These systems will advance the state of the art for utilizing biomedical data to improve healthcare decision making.  The personalization of medicine will allow for highly-effective treatment plans to be developed for patients based on their exact physiology, rather than simple statistics such as gender or age.

Flagship Biomedical AI: LITHIA

Our methods allow applying fuzzy logic based control to problems of extreme complexity. LITHIA, the LITHium Intelligent Agent, was created to apply our fuzzy logic based linguistic control to the prediction of effectiveness for Lithium treatments of Bipolar disorder patients.

Compared to other methods that can be applied to a problem of this complexity, LITHIA is incredibly accurate, robust, and transparent. Most alternative methods are similar to a black-box system in once cannot fully understand why the system provided a certain response. When asked why a patient received a certain prediction of effectiveness, LITHIA responds in an English if-then sentence. This allows LITHIA to be utilized a research tool, allowing medical doctors with little to no Artificial Intelligence expertise to work alongside LITHIA to determine key predictors and any variables that could be excluded.

On a 80% / 20% training / validation split of the patients, LITHIA is 100% accurate in predicting whether symptoms will be considered in remission after the treatment or not. The exact symptom reduction percentage is 94% accurate on this set of patient data.

8 other machine learning methods were tested, all varying between 50-75% accuracy in predicting symptom remission and 61-84% accuracy in exact symptom reduction percentage. These other methods also suffer from the black-box syndrome; their outputs provide extremely minimal insight into the problem compared to LITHIA.

It is clear why a genetic fuzzy system has not been applied to a problem such as this before. While the methodology has many strengths, a standard genetic fuzzy system would take much more than one googolplex years of computation time on the most powerful supercomputer in the United States. LITHIA can finish learning from the training data in under 5 hours, on a laptop. Post-training, running new patient data through LITHIA is nearly instantaneous.

Normal axial T2-weighted MR image of the brain