Anzyz deliver first AI-based coronavirus diagnostician to relieve pressure for COVID-19 test kits
On Monday, March 22nd, Anzyz Technologies released into laboratory testing the first AI-based COVID-19 diagnostic assistant.
The AI-based assistant, called “CARA” (Coronavirus Assessment and Research Assistant) was trained on 9,000 PubMed articles made available by the Allen Institute in Seattle, USA.
Specializing in coronaviruses and focusing on COVID-19, the web application is intended to be used by individuals who suspect they may have been infected, medical care providers as a preliminary step to diagnosis, and researchers.
CARA reduces the need for testing kits on patients who may not be carrying the Coronavirus and can screen a very large number of people quickly. The main difference between CARA and the other online diagnostic tools is that people can describe their symptoms with text, contra clicking on predefined symptoms that are given and that they feel do not apply to them. An example of this is “difficult in breathing” which people can experience and explain in various ways such as “chest ache”, “shortness of breath”, etc. CARA catches all of this and matches them correctly giving a more precise answer.
The CEO of the company, Svein Olaf Olsen, explains that “CARA is very easy to use. The AI is consulted by the user describing clinical symptoms in 10-12 sentences just like visiting a real doctor. Based on the symptom descriptions, hypotheses about a Covid-19 diagnosis is returned in seconds. The certainty of the diagnosis is probability rated by colour-coding. If red, you should immediately consult your doctor for a COVID-19 test. If yellow, you should continue to monitor your symptoms and their development. Green indicates that you do not have COVID-19.”
CARA is also an excellent Coronavirus research tool and can analyze hundreds of articles in seconds to extract and consolidate the information a researcher is likely seeking.
“While knowledgeable, CARA is only an assistant because we are still quite early in the development of General Artificial Intelligence,” said Olsen. ”Nevertheless, it was magical when we first consulted CARA. We asked it about COVID-19, and, after three minutes of questions, CARA told us about other Coronaviruses such as nl63, 229e and HkU1 as well as several others. In fact, by reading an article on nl63 we learned that this Coronavirus was first identified in a seven-month-old girl in Holland as early as 2004.”
Anzyz Technologies, based in Grimstad, Norway, is the developer of the Corpus Cube Linguistics (CCL) self-learning algorithm. Originally invented by its CIO, Professor Ole-Christoffer Granmo, and his technical team at the University of Agder, these natural language processing algorithms are used to train AIs that understand and operate in any language, dialect, or alphabet. They also automatically handle tasks that have been traditionally difficult for AIs to address, such as misspellings, humour, and sarcasm.
“We have even used CCL on Chinese. The system taught itself the Chinese language in 10 days,” says Olsen. “If we get access to patient records and other text-based information in any language, for example, Spanish, it will only take 10-12 days for CARA to be fully operational in that language.”
“This is a breakthrough technology and I am so proud of what Anzyz has achieved,” Olsen said. “When I introduced the system to an English health specialist today, he exclaimed; It's almost too good to be true!”