AI can detect Corona virus from CT scans in just 20 seconds
Alan Brown
Helping research scientists extract maximum value from data by applying artificial intelligence & autoML.
Chinese tech company Alibaba, says its AI can detect Corona virus from CT scans in just 20 seconds. It takes doctors around 20 mins to do the same. The model has an accuracy of around 96%.
The diagnosis algorithm was developed by Alibaba's research institute Damo Academy. Researchers at the academy said they had trained the AI model with sample data from more than 5,000 confirmed cases, adding that the system could identify differences in CT scans between patients infected with the novel virus and those with ordinary viral pneumonia with an accuracy of up to 96%.
The algorithm included the latest treatment guidelines and recently published research, said its creators.
Read more:
https://asia.nikkei.com/Spotlight/Coronavirus/Alibaba-says-AI-can-identify-coronavirus-infections-with-96-accuracy
CCIO, NT Health, AUS - Experienced Clinical & Digital Health Executive, Passionate about Human Centred Everything, GAICD
4yA few things here... 1) Hopefully we would be detecting COVID-19 well before a patient was unwell enough to justify having a CT performed. And in fact, what is the justification of performing a CT in acute pneumonia? It’s incredibly rare in usual intensive care practice at least. 2) the pretest probability of COVID-19 at the moment is so high, the real issue would actually be specificity, not sensitivity. 3) the lung changes of COVID-19 are non-specific and typical of other causes of ARDS. As a result I am not sure what problem this AI Is trying to solve, or even if it solves it particularly well. Any chance of more detailed information than a PR article?
Space Medicine trainee, Engineering student, Emergency Physician, Startup Founder, Mathematician, Programmer, Nerd
4yThere is *no way* any trained radiologist takes *20 minutes* to read a ct scan, more like 3-5 minutes I would assume based on my interactions with them
Sr. Machine Learning Engineer (MLOps) @ Tiger Analytics | AI & Space Enthusiast | Sustainable Interaction Design | Applied Data Science & Big Data @ WCD | Aerospace & Astronautics @ KingstonU
4yUsing accuracy as a performance measure for a highly imbalanced data-set (the ratio of positive cases to negative cases) is not a good idea to evaluate the AI’s performance, since the majority would be negative cases. Recall score is a better measure. Recall is calculated as the number of true positives divided by the total number of true positives and false negatives. Recall is intuitively the ability of the AI to find all the positive samples!
Application Development Consultant at The Goodyear Tire & Rubber Company
4yIn terms of problem being global, everyone should share their knowledge. In these circumstances, saying that research belongs to company X and not sharing research globally it's creating conspiracy and it doesn't help to solve problem. Alan Brown please share more than just article.
⚫Creator, World 1st Automobile for sustainable energy, and 3d humanoid friend Dio ⚫Inventor of "Supersymmetric Deep Learning" ⚫Manager of Ai at AICE ⚫Created math.../hypothesis: "Human purpose is likely to create AGI"
4yI foresaw this from February 9th, having started an ai based ct scan initiative: http://www.facebook.com/ProgrammingGodJordan%2Fposts%2F889038718221822