Articles, Blog

Computer Aided Diagnosis of Fungal Infections

August 14, 2019


>>Hi. My name is Eric Chang. This joined works with
Professor Yang Shi from Beihang University
and she is a fan. So today I’m talking
about Fungal Infections. Fungal Infection affect
more than one billion people a year, and kills more than 1.5 million
people a year worldwide. So it’s a very significant
healthcare problem. This was a joint project with Peking Union Medical
College Hospital, Beihang University, Pfizer,
and Microsoft Research. What I want to do is provide
a tool in the Cloud that allow doctors to help to make a diagnosis on those type
of Fungal Infections. So we’re using
a deep neural network called Bilinear CNN to help to
make the classification. With the Bilinear CNN, we increase the accuracy
from original 78-90 percent. So we’re trying to classify among five different classes
of fungal infections. So here’s the confusion matrix. You can see that we are enhancing the classification accuracy as
we get more data over time. So this is a typical image that maybe uploaded with system to generate
the classification result. What we’re doing is also providing
the activation map that’s used by the deep neural net
to make that determination. So this is very important
because there’s a new research called Printable-AI. One that give doctors
a way to sort of visualize how the deep
neural nets making decision. This was important so
that doctor can see whether the deep neural net is looking at the same features that
they were looking at. For example for the string
of fungal infection the color and the contrast at
the burp area is very important. So you can even see that the same feature is being
used by the deep neural net. So this another type
of fungal infection. So you can see that the symptoms also extracting all the features that
the doctor would be looking at. Also we cleaned up this
coefficient algorithm we provide the service in the “Cloud” that’s reachable through a mobile phone. With this mobile phone you
can upload information such as the image that you
see under the microscope, plus metadata such as the culture media that’s
used to grow the tempo, plus the conditions
such as temperature. Then the system will provide
suggested reference material to look at to help them make a better decision on what type
of a fungal infection it is. Also the system allows
the doctor to provide metadata such as additional
information about the patient, and also the treatment plan
and things like that. So one of the system provide
the reference material, he also provides a gallery. So unlike currently where the doctor only have
a reference books maybe with two or three images that they can use to look at the type
of a fungal infection. Now they have a constantly
enhanced and enlarging, they have said that they can look
at plus additional information such as doctors who provided the centerfold so they can actually
consult with other doctors. So we believe that overtime all professionals will use
tools such as these enhanced by AI to enhance their work
so that they can make better decisions over
time. Thank you very much.

No Comments

Leave a Reply