New breath test invokes machine learning to detect lung cancer

April 16, 2021

The new frontier in lung cancer detection involves the breath from those lungs. (AP Photo/Jacquelyn Martin)

A new test that detected lung cancer from exhaled breath with 96% accuracy is simpler and faster than similar breath tests, according to researchers, who suggest their advance is a step toward improved diagnostics for the prevalent disease.

In their diagnostic study, published March 30 in JAMA Network Open, researchers used a technique that analyzes a sample in just 60 seconds to develop a machine-learning model that accurately distinguished between healthy samples and those with lung cancer.

Worldwide, lung cancer is the second most commonly diagnosed cancer and causes the most deaths of any cancer, making it the focus of much research. In 2020, there were 2.2 million new lung cancer cases, and 1.8 million people died from the disease, according to the World Health Organization.

"There's no effective and simple method to detect lung cancer," said Mantang Qiu, a senior author of the study and a thoracic surgeon and research assistant at Peking University People's Hospital. "If we can develop a breath test, eventually I hope we can develop a test that can be used at home." 

Breath tests for detecting a range of cancers, including those of the lung, are in development. But the humidity of exhaled breath can complicate sample analysis, often requiring sample preparation and processing times of several hours. 

The current gold standard for diagnosis of lung cancer is low-dose computed tomography. But CT scanners are expensive and limited to certain clinics and can struggle to distinguish between cancerous and benign nodules, meaning that invasive procedures such as biopsies are needed to confirm suspected cancers, Qiu said. Breath tests, which measure the presence or concentration of gases called volatile organic compounds, hold promise as a rapid, cheap and noninvasive alternative for diagnosis.

For the study, Qiu and his colleagues recruited 289 healthy participants and 139 patients with lung cancer diagnosed by CT scan. After taking a deep breath, patients exhaled through the mouth into a handheld collection device connected to an air bag. One of the study's authors, Hang Li, has received a patent for this device issued by Shenzhen Breatha Biological Technology Company.

The bags of exhaled air samples were sent to a laboratory for analysis with a technique called high-pressure photon ionization time-of-flight mass spectrometry, or HPPI-TOFMS, which detects the individual chemicals in a sample. Unlike other methods for analyzing the chemical composition of samples, this technique is not affected by water vapor in breath samples and doesn't require pre-treatment, according to Qiu. 

The team used the results from 196 healthy and 90 lung cancer patients to train a machine-learning algorithm to discern differences between these samples. Then, they tested how well the model was able to classify 65 healthy and 30 cancer samples. Finally, they validated the model with an additional 47 samples, for which the researchers were blind to the correct diagnosis. 

Of these, the model accurately diagnosed all 19 lung cancer samples, for a sensitivity of 100%. Of the 28 healthy samples, the model accurately classified 26 and incorrectly classified two samples as lung cancer, giving a specificity of about 93%. Overall, these scores combine to give the model an accuracy of 96%.

"It's a preliminary study," Qiu said. "We showed that HPPI-TOFMS can detect the difference between lung cancer patients and healthy adults. But we don't know which volatile organic compounds are the key differences between the two groups." 

The next step is to identify the specific compounds associated with lung cancer patients, he explained.  

According to Qiu, another limitation of the study is that the researchers had a relatively small sample size and didn't validate the model with an independent dataset, such as patients from a different medical center. That is something that they plan to do in follow-up studies.

Also in future work, the researchers will examine differences in volatile organic compounds in patients with lung cancer compared with those with benign nodules, Qiu said. They additionally plan to investigate the use of breath tests for monitoring how lung cancer patients respond to chemotherapy or radiation treatment after surgery. 

"We want to see if breath tests can be a useful tool for monitoring disease recurrence," Qiu said.

The researchers' ultimate goal is to develop an at-home breath test, allowing people to regularly test themselves for lung cancer, Qiu said. He envisions that such a test might look something like an electric toothbrush and would ideally cost less than $50.

The study, "Assessment of an exhaled breath test using high-pressure photon ionization time-of-flight mass spectrometry to detect lung cancer," published in JAMA Network Open on March 30, was authored by Shushi Meng, Zuli Zhou, Xianping Liu, Mantang Qiu and Jun Wang, Peking University People's Hospital; Qingyun Li, Hang Li and Lei Wang, Shenzhen Breatha Biological Technology Co Ltd; and Shuli Pan, Yanqing Guo and Mingru Li, Aerospace 731 Hospital.

We use cookies to improve your experience on our site and to show you relevant advertising.