How AI And Deep Learning Are Now Used To Diagnose Cancer


Without a doubt one of the most exciting potential uses for AI (Artificial Intelligence) and in particular deep learning is in healthcare. Traditionally, diagnosis of killer illnesses such as cancer and heart disease have relied on examinations of x-rays and scans to spot early warning signs of developing problems.
 
Image recognition is of course one of the tasks at which deep learning excels – from Facebook’s facial recognition to Google’s image search, practical examples of it in use are becoming more common by the day.
 
Although being able to tag pictures of our friends without typing their name, or find amusing images of cats when we want them, may seem trivial use cases, the same technology is quickly advancing to a point where more far-reaching implications are being realized.
 
In China, lung cancer is the leading cause of death, claiming over 600,000 lives each year, largely due to high levels of air pollution. Radiologists work from CT scan images to hopefully diagnose sufferers at the earliest opportunity. But in a country where there is a serious shortage of qualified doctors, particularly radiologists, this often means they find themselves examining hundreds of images every day. It is incredibly tedious and due to fatigue, mistakes and misdiagnoses are not uncommon.
 
This was the problem that persuaded Chen Kuan, founder of startup Infervision, that medicine was the field in which he would focus his work with deep learning and image recognition.
 
Following a pilot project working with the Szechwan People’s Hospital, Infervision has now begun working with a number of the country’s top
 
Kuan told me “So what I saw was that a lot of Chinese people, particularly those living outside big cities, do not get to have any regular medical check-up involving medical imaging. So they often have to wait until they feel something wrong with their body before they go to a big hospital where it can be diagnosed.
 
“By then it’s often too late to do anything about it.
 
“So what we wanted to do is use deep learning to alleviate this huge problem. If we can use it to learn from the past and assist in diagnosing more accurately, we can help solve the problem.”
 
Kuan spent a year working with two other team members at the Szechwan hospital, in order to learn how the tool they were developing could be integrated with systems used in the hospital such as the Picture Archiving and Communication System (PACS). While there they were able to begin training their algorithms using real data in order to increase its accuracy at spotting warning signs of potentially cancerous nodule growth in lung tissue.
 
Deep learning involves the use of deep neural networks – algorithmic models designed to pass data along networks of nodes in a way which mimics the function of the human brain. These networks are able to adapt based on the data they are processing, as it passes through the network from node to node, in order to more efficiently process the next bit of data. Because of this they can be thought of as “learning” and able to teach themselves new ways of spotting danger signs.
 
The particular method employed by Kuan and his team is known as supervised learning, because data sets where the outcome is known were used to “teach” the model how to spot images which indicate danger. In this case this data would be previous CT scans which led to diagnosis of lung cancer.
 
“So basically, what we need, is a lot of data”, Kuan tells me. “And using that I managed to build a very simple model. Basically what I did was teach it to predict if an x-ray is normal or not. We know the healthy ones – so a radiologist now does not have to spend so much time on healthy ones and can focus more time on unhealthy ones. This is the foundation of what we are doing right now.”
 
In 2015 Infervision acquired investment and expanded its work to a number of other large hospitals in China. Now the company is seeking international partners to help relieve the workload of radiologists – as well as save lives – in other parts of the world.
 
This is an important factor that Kuan is keen to stress – that his company’s technology is not in any way meant to make human radiologists redundant, but assist them in diagnosing, and enable them to work with far greater accuracy and efficiency than has previously been possible.
 
“In China there are just 80,000 radiologists who have to work through 1.4 billion radiology scans every year. By using AI and deep learning, we can augment the work of those doctors. In no way will this technology ever replace doctors – it is intended to eliminate much of the highly repetitive work and empower them to work much faster.”
 
It’s certainly an exciting use case for AI and exactly the sort of work that we know machines are highly suited for, due to their ability to work until their power supply cuts out without ever suffering from a moment’s boredom or slip of concentration.
 
And with Infervision as well as other companies exploring AI-driven examination of medical images of many other parts of the body, I am confident we will hear more success stories like this very soon.