There’s plenty of buzz these days about artificial intelligence (AI), that magic potion believed to accelerate the power of any given system.

That’s the case in pulmonology. Currently, AI technologies are being used along with imaging to classify and screen disease, to identify lung abnormalities and to predict the need for preventative treatment, according to a recent story in Healio.

It’s also used in cytopathology, categorizing respiratory events in polysomnography and to assess lung physiology and respiratory function.

A recent study, published in Translational Lung Cancer Research, found that AI can help with bronchoscopy by enabling a quality-control system that improves the performance of bronchoscopists. The AI system, for example, was better able to recognize lumen than junior and senior doctors alone were, according to the authors.

“Our findings suggest that the supplemental application of the AI system could reduce the differences in the endoscopic skills of doctors with different levels of experience,” they added, “and it could become a powerful auxiliary tool to improve the quality of daily endoscopic examinations.”

It’s already proving valuable in endoscopy related to gastroenterology practice and research, according to a study published in Visceral Medicine.

What About AI Bias?

A National Institute of Standards and Technology publication, NIST Special Publication 1270, written by Lori A. Perine — the author of the Healio post and an industry expert — and others, identified three sources of AI bias: statistical or computational, human and systematic.

Those occur when there are:

• Issues with data quality or representativeness.
• Cognitive and/or social dimensions that influence decision-making and lead to systematic errors.
• Historical, societal or institutional biases that factor into data selection and model design decisions.

It is safe to assume that the pulmonology uses for AI may be subject to these biases, the NIST authors say.

“AI algorithms are trained on past data, and past data reflect the societal and institutional biases that are present when the data were collected,” Perine wrote.

Even decisions about which data to collect and which to use to train and test a model are subject to both explicit and implicit assumptions that incorporate human and systemic biases, she wrote.

Keeping It in Check

Among the ways pulmonologists can mitigate bias:

• Gain as much information as possible about data and assumptions used in model development.
• Identify geographic, demographic and socioeconomic differences in the presentation of a particular lung disease which may not be factored into the algorithm

“Once the limitations and potential biases of applying a particular AI algorithm are understood and documented, it is possible to employ debiasing or mitigation strategies to lessen disparate impacts for individuals, groups and communities,” Perine wrote.