Main menu


Artificial intelligence (AI) has begun to be used for early detection of cancer

Artificial intelligence (AI)  has begun to be used for early detection of cancer

Artificial intelligence (AI) has begun to be used for early detection of cancer-will neural networks change medical care?

It is no longer just the faces of humans and cats that artificial intelligence can distinguish from photographs. From the results of X-rays and CT scans, it is becoming possible to detect signs of cancer. In the not too distant future, neural networks may drastically change hospital operations, although they are still not as good as human doctors.

Wang Xiao Kang has launched a startup, Infervision , to read X-ray images and create algorithms to identify early signs of lung cancer. According to Wang, the company's technology has already been put to practical use in four of China's largest hospitals. Two of these hospitals are just testing, but the other two hospitals (Shanghai Choshu Clinic and Dousai Hospital in Shanghai) have fully implemented the technology. "This algorithm is installed on every doctor's machine," Wang says.

Of course, how much doctors actually use this technique is another matter. The idea of ​​introducing artificial intelligence into the medical world is still in its infancy, but it is still spreading.

At two hospitals in India, Google is currently testing a technique to identify signs of diabetic blindness by scanning the eyeballs. In May 2017, data science competition site Kaggle announced the winner of a contest to build a machine learning model to detect lung cancer with CT scans. More than 10,000 researchers participated in the $ 1 million prize contest. The winning algorithm reads the work of the National Cancer Institute and diagnoses lung cancer more quickly and effectively. "We want to further evolve these solutions," said Kayvan Farahani, program director at the institute.

Not only is it difficult to introduce this technology into existing systems and daily work, but it is also extremely difficult to aggregate all the necessary data. However, Sea believes that the best algorithms currently being made are already accurate enough to withstand commercialization. "Probably it will only take a few years before it is introduced on a larger scale," says Sea.

Between medicine and computer science

These systems are the result of deep neural networks, a complex system that can automatically learn tasks by analyzing vast amounts of data. Originally an old idea invented in the 1950s, neural networks can do more than ever, as companies like Google and Facebook can now handle vast amounts of data and computing power. Can be achieved. In particular, it can accurately recognize faces and objects in photographs. In addition, scanning medical data can identify signs of illness or illness.
Neural networks can identify small aneurysms with eye scans or tiny nodules with lung CT scans, just as cats can be identified from living room snapshots. Basically, after analyzing thousands of images containing such nodules, neural networks are learning how to identify them on their own. Thousands of data scientists competed to build the most accurate neural network for this task through Kaggle's contest.

In order for neural networks to start learning from a database of images, trained doctors need to label those images. In other words, it uses human intelligence and knowledge to identify images that show signs of cancer. Once that work is done, the construction of these systems is in the realm of computer science rather than medicine. A good example is that two Kaggle award-winning researchers at Tsinghua University in China, Liao Fanjo and Lee Ji, did not study medicine properly.

AI as a doctor's assistant

Nonetheless, these AI techniques do not completely replace trained doctors. "We're still doing only a small part of what radiologists and doctors are doing," says Sea. "There are still dozens of other illnesses that we need to unravel." The new AI system scans faster and more accurately before doctors take a closer look at the patient's situation. These AI-powered assistants would ideally reduce medical costs. Doctors have to spend a great deal of time on screening, and doctors can make mistakes.
According to Shee et al., Doctors rarely make false negative decisions. However, it is a problem to make a positive diagnosis by mistake. Hospitals often spend time and money tracking the progress of patients who do not require as much care. "The problem with lung cancer screening is that it's very expensive," says Sea. "The big goal is how to minimize that cost."

Sea's company aims to build a service that collects and labels data that can be used for training neural networks for tasks other than cancer detection. She admits that the development of this kind of AI is still in its infancy. But he believes AI will radically change the field of healthcare, especially in developing countries where there aren't many trained doctors yet. According to Sea, researchers are unlikely to build an AI that is better at detecting lung cancer than best-in-class doctors in the next few years. But even if the machine outperforms some doctors, even if it changes the way the hospital operates, it will only be scanned once.