18 Oct Artificial intelligence and healthcare technology
Brookfield, Wis. – An artificial intelligence (AI) technology developed by a southeast Wisconsin firm has proved in initial testing that it can make diagnoses based on pathology samples.
According to Tom Keeley, president of Brookfield-based Compsim, LLC and a Houston-based physician conducting research using Keeley’s Knowledge Enhanced Electronic Logic (KEEL) tool, initial diagnostic tests have been successful.
Compsim’s AI tool was able to diagnose conditions based on samples after being “taught” to do so by an expert physician. KEEL originally was an offshoot of a software product Keeley developed to help management teams make better decisions and document how they arrived at a course of action.
But he realized a portion of the program code that emulated the human thinking process had potential by itself. When matched with a set of weighted criteria provided by a human expert, Keeley would have a system for embedding human thinking and knowledge in devices. Since then, Keeley has been seeking licensing deals with companies that might be interested in embedding intelligence in new or existing products.
“It went pretty well – like traditional interpretation,” said Jeff Chang, a professor with Baylor College of Medicine, Houston. Chang said the challenges he encountered had primarily to do with differences between different makes of automatic analyzer equipment. Results could differ from one machine to the next, and the KEEL logic engine had to be reprogrammed accordingly.
Chang programmed KEEL to analyze the samples, in effect digitizing his own thought processes and expertise, “teaching” the software to think like he did. KEEL made diagnoses on 125 samples, according to Chang.
“I taught the instrument how to do an interpretation,” Chang said. “I gave the software my thinking process and how I analyze the samples through the Keel engine.”
Chang started his own company – 4E Biochemical – to develop medical applications for Keeley’s AI tool. He and Keeley hope to be able to embed KEEL intelligence into equipment that currently provides hard data analysis of pathology samples but does not interpret the data and provide diagnoses.
Different kind of intelligence
According to Keeley, the differences between KEEL and other AI tools including fuzzy logic and neural networks make his technology a logical choice for medical applications like the ones Chang is exploring. Fuzzy logic is a computational system that uses approximate rather than precise information for decision generation. Neural networks seek to imitate the human brain through the use of highly interconnected processing elements tied together with weighted connections similar to synapses.
KEEL, however, uses a simple graphical code consuming about 2K of memory to make decisions based on criteria entered into tables. The small size means KEEL could be easily embedded in a number of devices. Like a neural network, KEEL uses weighting factors to prioritize criteria. But unlike a neural net, the KEEL programming engine allows skilled people to program in their own decision-making criteria and tweak the system until KEEL’s decision-making matches their own.
John Byrnes, executive managing director of Milwaukee private equity investment firm Mason Wells, said Compsim’s technology has a place in the market.“Their technology is a form of AI that goes along with other types like neural networks and fuzzy logic – and I think it has potential for being very useful in a number of areas,” Byrnes said.
Byrnes said the he and Keeley both think that highly skilled people do not make decisions intuitively, but rather use highly-complex metasystems that are explainable and can be documented.
“It has a human interface that allows it to be tuned by an expert,” Byrnes said. “You can tune the system with the input of a professional who might have his own metasystem. The whole idea that human beings use metaystems to make judgments is correct.”
The fact that decisions made by KEEL are explainable and can be documented after the fact is significant as well, according to Keeley.
“In the case of these pathology samples, we will be able to automate reports that not only tell what condition someone is likely to have, but explain why the determination was made and why other diagnoses are not consistent with the test results,” Keeley said. “That has applications for communicating to patients and to insurance companies.”
Other applications pending
Chang said that apart from pathology, the KEEL tool might have medical applications in research.
“In microarray research, those tests are so big, no one knows how to analyze it,” Chang said. Microarray research involves looking for specific genes in a human cell. These genes cause genetic alterations resulting in formation of cancer cells. Because the human genome contains about 100,000 genes, and a single cell may express up to 15,000 to 20,000 of these genes simultaneously, automation of the analysis process would be beneficial, according to Chang.
Keeley is also pursuing other applications in other industries, including the automotive industry and the military. He visited this summer with an Army command in Orlando, Fla., that connects military sources with emerging technology. Keeley said he pitched his AI tool as a way to potentially reduce manpower devoted to maintenance as equipment could be programmed to diagnose and troubleshoot itself.
But Byrnes, who has introduced several companies to what Keeley is up to, said licensing arrangements like the ones Keely is seeking can be hard to nail down and difficult to manage.
“The challenge I think he has is that it is not a product in itself,” Byrnes said. “It has to be embedded in a product. Keeley’s strategy is to seek licensing deals. Licensing strategies have had limited success over the years. Most businesses want their products to be proprietary, and they wouldn’t want KEEL technology everywhere.”
Another challenge faced by Compsim, according to Byrnes, is their location in Wisconsin. “Unfortunately for Tom, the type of community that really understands that technology doesn’t really exist here in the state,” Byrnes said. “So he has been forced to go other places to try to find people to work with. If aggressively employed, KEEL could find its way into several applications. What we need in Wisconsin is companies that want to develop intelligent, adaptive applications.”
Charles Rathmann is a freelance writer and contributor to Wisconsin Technology Network. He can be reached at firstname.lastname@example.org.