Lung cancer has long been one of the most difficult forms of the disease to diagnose.
Doctors typically use their eyes to examine CT scan images, looking for small nodules in an attempt to deduce whether they’re benign or malignant. When the nodules are small, they’re harder to spot. And the result is that lung cancer is often detected too late, leading to a dismal 17 percent survival rate, according to Xin Zhong, co-founder and CEO of 12 Sigma.
“Almost every single lung cancer started as a small nodule,” said Zhong. “That nodule can have a variety of looks, and it takes doctors years to know all the different looks.”
Better Cancer Diagnosis with Deep Learning
Zhong and his team decided to use deep learning to train an AI algorithm that would help doctors analyze CT scan images more efficiently, resulting in quicker and more accurate lung cancer diagnoses. CT scans consist of up to 500 images. As these and other image-based medical tests have become more common, so has the need for powerful analytical technologies.
“AI is suited for detecting cancers because there has been a vast amount of data being produced,” said Dashan Gao, co-founder and CTO of 12 Sigma. “We can use the latest deep learning computer vision technology to process the data.”
The company uses CT images from hospitals in the U.S. and China to train its models on GPU-powered neural networks that run 50 times faster than those running CPUs. Gao said that’s translated to being able to train models in days or even hours instead of weeks or months.
Constant Learning with CT Imagery
12 Sigma’s technology is being tested in at least 35 hospitals in China, where doctors typically take a first pass on CT images using their eyes, then run them through the 12 Sigma algorithm. They often find cancer nodules that they missed during the eye test, resulting in doctors making quicker decisions to remove malignant nodules.
The technology keeps learning as doctors remove false positives and use false negatives, establishing newly labeled data and fine-tuning the algorithm.
“By working with the doctor and the software together, we can achieve very high sensitivity and very low false positives,” said Gao.
A Literal Life-Saver
The impact on doctors has been significant. Previously, it took doctors as long as 10 minutes to visually inspect a CT scan image. Additional time is needed classify them as benign or malignant. The 12 Sigma system takes two minutes, including classification. Zhong estimates this saves doctors at least four hours a day, freeing them up to communicate with patients or to pursue more research.
Of course, the real victory is in keeping patients alive by identifying lung cancer sooner.
“Saving someone’s life is meaningful,” said Yumin Yuan, vice president of product development for 12 Sigma. “That’s what we’re excited about.”
The next episode in our “I Am AI” docuseries will feature ModiFace, which is applying its groundbreaking beauty industry AI to help people find their ideal hair colors.
Watch more episodes of the “I am AI” docuseries on our AI Innovators page.