AI Case Study
Beijing Tian Tan Hospital is testing the detection of type, location and severity of a stroke using machine learning
Beijing Tian Tan Hospital is testing the detection of type, location and severity of a stroke using a machine learning from Infervision.
Accurately estimating blood loss and location of a stroke is critical to determining treatment. Time is critical due to brain cell death after a stroke. Training deep neural networks on CT images of brain stroke patients allows diagnosis to be made more quickly and accurately.
Healthcare Providers And Services
Tian Tan Hospital radiologists are piloting a solution from Infervision, a leading Chinese medical image diagnostic company.
"To develop this haemorrhage volume diagnostic capability, the Infervision platform applied deep learning technology and trained many thousands of datasets of annotated medical images."
"With the Infervision platform, doctors may take scans with the much more readily available CT machine, and use the AI technology to help guide them in making a faster diagnosis and perhaps save more brain tissue with faster and more appropriate treatment."
Strokes are the "third leading cause of death and the leading cause of permanent disability and loss of independent life-years in Western countries."
Stroke “'haemorrhage volume is strongly associated with mortality and the best way to intervene'. Volumes over 30ml are strongly associated with mortality and its better to use aggressive surgical methods to intervene. The problem is, during our testing phase we asked radiologists to conduct these calculations and we found that in some cases the margin of error was more than 30ml.”
When someone suffers a stroke it is critical to be able to measure the blood loss and location accurately. But often it is something of a finger in the air. And every minute that goes by can impact the patient outcome as brain tissue cells die.
"For hemorrhagic stroke patients, the AI-CT Stroke Screening System technology assists doctors in accurately and quickly determining whether a bleeding-type stroke has occurred, how much blood volume is involved (which is quite difficult and often inaccurately estimated through other methods), and the bleed location - crucial factors in deciding treatment options."
Deep supervised learning. Likely convolutional neural networks.
"Over 100,000 annotated medical image scans were used to train the algorithms, which given more live data will become increasingly efficient at diagnosing the two main types of stroke, hemorrhagic and ischemic."