The field of health and wellness is one of the most important applications of STEM knowledge that people face. We have developed a proprietary artificial intelligence system (AI) based on mathematical constructs with a medical diagnostic knowledge base using our Argo™ Clinical Decision Support System (CDSS). We have intentionally stayed away from medical therapies, except for their side effects, in order to avoid the appearance of practicing medicine.
Medical AI systems are notorious for being difficult to maintain because of the complexity of the subject matter and the complexity of the computer algorithms used to represent knowledge. More recently, neural networks (machine learning) and big data sets have used "brute-force" computation to make AI systems more human-like. The AI herein does not try to mimic humans but attempts to use mathematical structures to make connections between causes and effects. Argo algorithms use relatively straight forward mathematical concepts to measure similarities among medical conditions (causes) based on clinical measurements (effects).
The footprint of this system is very small and was designed to run on an original IBM-PC with 256 MB of RAM. It could easily be embedded in a mobile device, possibly for miliary or space travel uses.
The current contents of the knowledge base has approximately 6000 medical conditions, 5000 clinical tests, and 4000 drug reactions and toxicities with over 1500 symptoms and signs. The completeness of this content is only limited by the resources needed to enter the knowledge elements and medical content stored in a proprietary knowledge representation. Maintenance of the knowledge base is quite simple relative to rule-based or other inference engines and does not require massive computing systems or large databases.
For the Argo inference engine, Descipher has created a proprietary algebra for adding the effects of multiple medical conditions. This allows existing conditions and drug effects to be filtered out when seeking a new diagnosis. The inferences are not based on data, but on knowledge elements from the published peer-reviewed medical literature. However, actual locally data-driven probabilities and data-derived correlation structures could be added to the existing inference algorithm. The use of such data makes the inferences amenable to machine learning, but not in the neural-network sense.
The theory of pathodynamics, motivated by combining thermodynamics and financial mathematics, could be applied to this system with some effort. This would allow for the use of dynamic patterns of the biological systems to make inferences. Additional references are included in the Bibiolography section of the About Us page.