Google and UCSF researchers have developed an AI system that can predict doctors' prescription decisions in 75 percent of cases, according to Lei Feng. If applied to a medical system, it can identify patients or prescriptions that appear to be abnormal. It is similar to the fraud detection scheme used by information card companies.
and this result is published in the journal clinical pharmacology and therapeutics (Clinical Pharmacology and Therapeutics).
Research scientist Catherine
For this purpose, the AI system uses a data set for training, which contains about 3 million drug orders generated by more than 100000 inpatients, through the use of traceable electronic health records with randomly changed dates, and delete some records (including name, address, contact information, record number, doctor name, free text annotation, image, etc.) according to HIPAA. More importantly, data sets are not limited to specific diseases or medical fields, making tasks more challenging, but also helping to ensure that models can identify a wider range of disease types.
researchers evaluated two models: learned short- and long-term memory (LSTM) recurrent neural networks that model long-term dependence, and similar logical models commonly used in clinical health studies. both were compared with the baseline, which ranked the most commonly used drugs according to the patient's hospital services (e.g., general medical, general surgery, obstetrics, cardiology) and time since admission.
Each time a drug is ordered in retrospective data, these models list 990 possible drugs. Researchers evaluate these models and decide whether to assign the drugs actually ordered in each case to doctors with a high probability. The performance of each model was evaluated by comparing its ranking selection with the drugs actually prescribed by doctors.
Best performance is LSTM
The researchers wrote,
I have to say that Google's work on AI medicine is very extensive.
Lei Feng net (public number: Lei Feng net) learned that Google had previously developed a model that could be used
one recent google achievement is the use of AI models to identify breast cancer from breast X light films with fewer false positives. Google is also working with Alawende Eye Hospital (Aravind Eye Hospital) in Madurai, India, to diagnose eye disease from retinal images.