Source: ATYUN AI platform
At the hospital, doctors and nurses will be closely monitoring the patient's vital signs and blood tests to find out the first symptoms of sepsis. In this life-threatening disease, the body will cause a variety of infections inflammation of organ failure to respond. Cases can rapidly develop into severe sepsis and septic shock development, in this case the US mortality rate close to 50%.
But even the most vigilant people will feel tired, make mistakes and miss subtle change mode. This is the cause of several hospitals are experimenting with AI sepsis detector. The researchers said the pilot project is the first real example of AI be included in hospital operations, electronic medical records data and alerts into the doctor's workflow.
Next month, located in Durham, NC Duke University Hospital officially launched Sepsis Watch, which is an AI-based system for identifying early cases of sepsis and trigger an alarm. Hospital initially be deployed in the emergency room, then extend it to the general hospital and intensive care unit. "The most important thing is to be in front in the case of early detection, that they enter the ICU," one of innovation and project leader Duke Institutes of Health Suresh Balu representation.
Sepsis monitored by deep learning training, to identify cases based on dozens of variables, including vital signs, lab results and medical history, their training data including 50,000 patient records, including more than 32 million data points. In operation, every 5 minutes to extract it from the patient's medical record information to assess their situation, provide intensive real-time analysis of human doctors can not provide. If the system determines that the standard AI eligible patients with sepsis patients whose symptoms of early, rapid response nursing team will alert the hospital.
Doctors at Duke Institute and data scientist Mark Sendak said that now AI can not do everything. When she arrived at the patient's bedside, their job is to decide whether to lift the alarm, the patient is placed on a watch list, or discuss with your doctor to begin treatment. If there is a guide, sepsis observation system will also guide the staff to complete a project called "Save sepsis" treatment list of steps the global activities of the proposed plan, including blood tests and drug therapy within the first three hours. "The model detects sepsis," Sendak said, "but most applications are focused on the completion of treatment."
Sendak said the team carefully consider the user interface and alert system how to adapt to existing workflows. Clinicians increase their bout interference cautious: Sendak said, Duke Hospital in 2015 to try different times for an early warning system to identify sepsis cases will be issued 100 alerts a day to a patient.
Duke system is not the first application of AI in hospital sepsis detector. Early warning system that honor belongs to the implementation of the University of Pennsylvania Hospital, the hospital associate professor Craig Umscheid explained. His team at the beginning of 2016 to start the system, and in 2017 to turn it off. Umscheid said that the system did not improve the quality of care or patient outcomes, partly because when it identified potential sepsis patients, and medical personnel saw it long. "Identifying unexpected cases the opportunity to lower than expected."
Assistant professor of computer science at Johns Hopkins University Suchi Saria said at Johns Hopkins Hospital in Baltimore, similar systems show better results. Saria's team at the end of 2017 launched the AI system, she said it worked well, so that they are ready to extend it to other four hospitals. "We see care has undergone major changes, sudden worsening of the patient cases has fallen." Hopkins sepsis detectors tailored for different patient groups, for example, it evaluates compromised immune systems based on different criteria of patients, and also it has a workflow optimized for each unit of the hospital.
Sendak Duke University said that if these AI systems to improve care, many hospitals are eager to adopt the technology. From July 2018, the US government hospital comparison site began publishing data on hospital records provide early and adequate treatment for sepsis. "National average of about 50%, many places are trying to solve this problem."
This switched ATYUN artificial intelligence media platforms, the original link: AI system to recognize sepsis early and reduce mortality
more recommendations
AI algorithms help in unknown clutter UAV autonomous navigation environment
Python accelerate data analysis of 10 simple tips (on)
Argo AI and Carnegie Mellon University jointly set up autonomous vehicles research center
