25.04.2018 Views

Innovation i tech High

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

DIGITAL ECOSYSTEM<br />

The Solution and the<br />

Process<br />

We have a clear vision:<br />

to reinvent cost effective<br />

disease management<br />

through early detection<br />

and intervention. We are<br />

tackling this long term goal<br />

by making cancer screening<br />

and diagnostics as accurate<br />

and accessible as possible,<br />

by using advanced analytics<br />

tools like image processing<br />

and deep learning, Natural<br />

Language Processing (NLP)<br />

to improve patient care,<br />

hospital administration,<br />

supply chain and logistic<br />

efficiencies.<br />

The process is to first digitize<br />

the sample slides through<br />

a microscope, scan whole<br />

slides and upload onto the<br />

cloud by converting the large<br />

image into a binary metadata.<br />

AI engine then takes over,<br />

which classify and tag the<br />

visual data, not only to churn<br />

out results, but also support<br />

it with visual evidence (like<br />

tagging & annotations) to<br />

make it easily verifiable.<br />

With the advancement of<br />

AI, as more and more new<br />

data keeps coming in, the<br />

machine continues to learn<br />

itself like humans. And it<br />

starts generating solutions.<br />

There is no need for a manual<br />

review by a pathologist<br />

by putting a slide under<br />

a microscope. Remote<br />

diagnosis is made possible,<br />

because the pathologist<br />

need not be sitting next to<br />

the microscope or blood<br />

slide. They can access from<br />

anywhere, anytime.<br />

Our method is primarily<br />

based on a convolutional<br />

neural network (CNN) and<br />

the Wide Res Net 50 residual<br />

network formulation. In our<br />

approach, we first trained the<br />

CNNs on a suitable set of test<br />

datasets. Then we applied<br />

the trained deep model to<br />

partially overlapping patches<br />

from each whole slide image<br />

(WSI) to create prediction<br />

heatmaps.<br />

By utilising machine learning<br />

on whole slide images<br />

(WSI), we have developed<br />

a prototype for automatic<br />

detection of metastasis in a<br />

lymph node. Our early results<br />

of detection were shown to<br />

have an accuracy of upto 73<br />

percent.<br />

Our deep learning expert team<br />

and a pathologist is needed<br />

to work closely to translate the<br />

deep learning algorithms to<br />

digital pathology tasks.<br />

It is not that AI will entirely<br />

replace histopathologists, but<br />

rather it will provide increased<br />

efficiency and accuracy to<br />

the diagnosis, by providing a<br />

preliminary diagnosis much<br />

faster.<br />

About the Author<br />

Khalid Shaikh is the Founder,<br />

CEO of Prognica and Affaan<br />

Technologies. An Entrepreneur,<br />

a Technocrat and a Business<br />

Strategist, he is passionate<br />

about disruptive ideas that<br />

transform the human experience<br />

and create meaningful lasting<br />

change. Parallel to this, he<br />

has been involved in few <strong>tech</strong><br />

startups as an advisor and<br />

mentor.<br />

50 www.innovationand<strong>tech</strong>.ae<br />

INNOVATIONANDTECH<br />

January | 2018<br />

47

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!