Artificial Intelligence in Cancer: How Is It Used in Practice?

Artificial intelligence (AI) comprises a type of computer science that develops entities, such as software programs, that can intelligently perform tasks or make decisions.¹ The development and use of AI in health care is not new; the first ideas that created the foundation of AI were documented in 1956, and automated clinical tools that were developed between the 1970s and 1990s are now in routine use. These tools, such as the automated interpretation of electrocardiograms, may seem simple, but are considered AI.

Today, AI is being harnessed to help with “big” problems in medicine — such as processing and interpreting large amounts of data in research and in clinical settings, including reading imaging or results from broad genetic-testing panels.¹ In oncology, AI is not yet being used broadly, but its use is being studied in several areas.

Screening and Diagnosis

There are several AI platforms approved by the US Food and Drug Administration (FDA) to assist in the evaluation of medical imaging, including for identifying suspicious lesions that may be cancer.² Some platforms help to visualize and manipulate images from magnetic resonance imaging (MRI) or computed tomography (CT) and flag suspicious areas. For example, there are several AI platforms for evaluating mammography images and, in some cases, help to diagnose breast abnormalities. There is also an AI platform that helps to analyze lung nodules in individuals who are being screened for lung cancer.^1,3

AI is also being studied in other areas of cancer screening and diagnosis. In dermatology, skin lesions are biopsied based on a dermatologist’s or primary care provider’s assessment of the appearance of the lesion.¹ Studies are evaluating the use of AI to either supplement or replace the work of the clinician, with the ultimate goal of making the overall process more efficient. 

Big Data

As technology has improved, we now have the ability to create a vast amount of data. This highlights a challenge — individuals have limited capabilities to assess large chunks of data and identify meaningful patterns. AI is being developed and used to help mine these data for important findings, process and condense the information the data represent, and look for meaningful patterns.

Such tools would be useful in the research setting, as scientists look for novel targets for new anticancer therapies or to further their understanding of underlying disease processes. AI would also be useful in the clinical setting, especially now that electronic health records are being used and real-world data are being generated from patients.

Another aspect of big data is the information produced by wearables and within social media channels.⁴ Depending on the device and applications used by the individual, wearables can generate both behavioral and physiologic data that could be mined for insight on interventions — such as modifying behavior to reduce the risk of cancer or identifying adverse events.

A current limitation for wearables is their accuracy in measuring physiologic variables. Social media also generates a vast amount of behavioral data, and studies suggest that the types of posts that are made, the language and tone that is used, and the searches that are performed can also be mined by AI to identify people who may be experiencing symptoms related to a medical condition.

Clinical Decision Tools

As the understanding of cancer continues to grow, and numerous new therapies are being approved, the complexity of cancer management has dramatically increased.⁵ As a result, AI is being developed to support clinicians in their decision making, with the goal of improving quality of care.^5,6 This may require the use of big data to identify meaningful patterns or to help researchers come to conclusions, but the use of those data would be to create a tool that would help clinicians make decisions about cancer treatment.

An example of this is an AI platform that provides clinical decision-making support by making treatment suggestions.⁷ The platform bases its suggestions on clinical practice guideline recommendations, analysis of the scientific literature, learning from experts and test cases, and the patient’s characteristics. An early study of this platform showed that the AI technology chose treatments that were highly concordant with what prostate cancer specialists would select.

In addition, electronic health records and other databases or registries are rich sources of real-world data.⁶ AI analysis of these data can be used to help clinicians and their institutions make better decisions. For example, the American Society of Clinical Oncology developed CancerLinQ, which houses electronic health record data from participating institutions, and then uses AI to analyze these data — data that members of the industry group can access for research purposes.^6,7

Conclusions

There are many potential applications of AI that are being developed to support oncologists in the areas of cancer screening and diagnosis, processing and analyzing big data, and clinical decision making. AI platforms are already in use in some areas of oncology, including for support in the screening and diagnosis of cancer, for the identification of biomarkers or treatment trends, and for the evaluation of large databases.

References

1. Yu K-H, Beam AL, Kohane IS. Artificial intelligence in healthcare. Nat Biomed Eng. 2018;2(10):719-731.
2. Data Science Institute. American College of Radiology. FDA cleared AI algorithms. Accessed May 5, 2020.
3. Arterys Inc. Lung AI website. https://arterys.com/lung-ai/. Accessed May 5, 2020.
4. Wang F, Preininger A. AI in health: state of the art, challenges, and future directions. Yearb Med Inform. 2019;28(1):16-26.
5. Yu SH, Kim MS, Chung HS, et al. Early experience with Watson for Oncology: a clinical decision‑support system for prostate cancer treatment recommendations [published online April 25, 2020]. World J Urol. doi: 10.1007/s00345-020-03214-y
6. Kantarjian H, Yu PP. Artificial intelligence, big data, and cancer. JAMA Oncol. 2015;5:573-574.
7. American Society of Clinical Oncology. ASCO CancerLinQ. Website. https://www.cancerlinq.org/solutions/oncology-practices. Accessed May 5, 2020.