AI in Oncology: Current and Future Applications

Using artificial intelligence (AI) in oncology has shifted from prospect to reality in recent years, and ongoing studies suggest AI could have many additional applications in cancer care. 

“Most of the AI that is currently being used in the practice of oncology is in the realm of cancer diagnostics,” said Tufia C. Haddad, MD, of Mayo Clinic in Rochester, Minnesota.¹ “Machine learning and deep learning models are being utilized to improve the accuracy and efficiency of making a cancer diagnosis.”

A plethora of AI technologies have received approval from the US Food and Drug Administration (FDA) for use in oncology, most notably in radiology.^2,3 In 2021, Luchini et al reported that AI-associated devices approved for use in oncology settings were most commonly used in radiology (54.9%) and pathology (19.7%), and the devices were used for breast cancer (31.0%) more often than other cancers.³

Breast radiologists are increasingly using AI to “assist with tasks such as determining breast density, determining the quality of a mammogram, triaging mammograms into those at low, intermediate, or elevated risk of having breast cancer, identifying those at risk for atherosclerotic disease, and identifying breast cancers,” said Laurie Margolies, MD, of Mount Sinai in New York, New York.^4,5

“AI can also be used with breast ultrasound exams to provide an additional piece of information — that is, whether the computer algorithm thinks a breast ultrasound finding is likely to be a cancer,” Dr Margolies added.⁶   

AI has also proven effective for the screening and diagnosis of colorectal cancer.⁷ In 2021, the FDA authorized marketing for GI Genius, which was the first device that used AI to aid the detection of lesions during a colonoscopy.⁸

In addition, AI is being used in some pathology labs to read digital pathology slides and refine cancer diagnoses, after years of research in this area, according to Olivier Elemento, PhD, of Weill Cornell Medicine in New York, New York.⁹

Dr Elemento also noted that AI is being used in the development of new cancer drugs.

“Some companies are using AI to discover novel and safe targets in oncology, and other companies are using AI — sometimes combined with molecular modeling — to design entirely new drug candidate molecules,” Dr Elemento said.^10,11 “The future of AI in drug discovery is very bright, in my view.”

AI is also being used to “improve the efficiency of radiation treatment planning and assisting with tumor and organ contouring, thus increasing the speed to therapy and improving the effectiveness and safety of radiation delivery,” Dr Haddad said.¹²

“For supportive care, AI is being used to assess remotely monitored patients’ self-reported symptoms and vital signs,” she added.³ “NLP/ML [natural language processing/machine learning] models can trigger alerts to adverse health trends.”

Applications Under Investigation

A range of additional applications of AI in oncology are showing promise in ongoing research.

“Breast imaging is ripe for AI research, and future applications might be able to identify near- and long-term breast cancer risk, perhaps better than family history or as a tool to be used in conjunction with family history,” Dr Margolies said.

Researchers are also exploring the role of AI in identifying patients with a high risk of pancreatic cancer, using abdominal imaging and longitudinal patient electronic health records, according to Dr Haddad.¹³

“This may enable screening programs for high-risk individuals to facilitate early detection in a disease associated with high mortality rates,” she said.

Dr Haddad added that AI could ultimately transform biomarker assessment and molecular characterization of cancers to inform prognosis and treatment decisions, thus obviating the need for invasive biopsies and the long wait for results in some cases.

“Imagine this being made possible through a blood draw or a noninvasive ‘virtual biopsy’ made possible by using MRI radiomics; for example, a deep learning model could evaluate a brain MRI and determine if a brain tumor has a specific genetic mutation, such as IDH1 or BRAF,” she said.

Dr Elemento described large language models (LLMs) as an exciting area with the potential for applications in oncology.

“There is anecdotal evidence that some of the existing OpenAI models may already be able to answer medical questions with good accuracy, but this needs to be validated more systematically using high-quality ground truth datasets,” Dr Elemento said.

“I think medical centers are sitting on petabytes of medical oncology data that could soon be used to train high-quality LLMs capable of ingesting lab results, scans, and prior medical history to make highly personalized treatment recommendations.”

In a study published earlier this year, Haver et al found that ChatGPT generated accurate responses to 88% of questions regarding screening and prevention in breast cancer, as confirmed by breast radiologists.¹⁴

A pair of studies published in August showed that chatbots can answer questions about a variety of cancers with high accuracy, but the technology still has limitations.

In one of these studies, Pan et al assessed chatbots’ responses to the top Internet searches related to skin, colorectal, prostate, lung, and breast cancers.⁵ The chatbots generally provided high quality information, but it was not always actionable, and it was written at a college reading level.

In the other study, Chen et al found that a chatbot’s responses to queries about cancer treatments did not always align with recommendations in National Comprehensive Cancer Network (NCCN) guidelines.¹⁶ The chatbot was able to provide at least 1 treatment recommendation for 98% of queries about breast, prostate, and lung cancer, and all of these responses included at least 1 NCCN-concordant recommendation. However, 34.3% also included at least 1 non-concordant recommendation, and 12.5% of the answers were “hallucinated” (ie, not part of any recommended treatment).

Yet another application of AI that is under investigation is the use of natural language processing models to predict survival in cancer patients. Nunez et al recently showed that these models can predict survival outcomes based solely on data from a patient’s initial oncology consultation.¹⁷

Optimizing AI for Clinical Use

Multiple limitations and issues must be addressed to help optimize the use of AI in clinical oncology, according to experts.

Dr Haddad emphasized the ethical imperative to develop AI models and AI-enabled medical devices with diverse data sets to ensure accurate representation of the “patient cohorts to which they will be applied when implemented in clinical practice. Without this, the models and devices can introduce bias and exacerbate health care disparities.”¹ 

One barrier to using large and diverse datasets in research is the hesitancy of medical centers to share data, Dr Elemento said.

“We either need to figure out ways to normalize and incentivize data sharing, ideally with support from patients, or come up with new training algorithms that can train AI models locally and merge the models across different centers and then verify the accuracy of the merged models,” he said.

Despite all the progress and the potential for the wider use of AI in medicine, many existing models remain to be assessed in clinical practice.

“An AI model is only a model if it sits on a shelf,” Dr Haddad said. “True AI in oncology is a model that is used in clinical care, provides value to clinicians, and improves patient and health system outcomes. The usability, acceptance, and safety of these models need to be defined, and ongoing evaluation will be needed to understand their impact on outcomes.”

Dr Elemento cited implementation challenges that currently limit the adoption of AI models in the clinical setting, including difficulty deploying AI models into electronic health record software, the lack of intuitive user interfaces in many AI models, and the prohibitive cost of training some of the recent models.

“This trend, if not addressed, will severely limit competition and access,” he said. “We either need to subsidize computing power or come up with more efficient training strategies that use less resources.”

Dr Haddad also highlighted the need for “guidelines and guardrails” to ensure the safety of AI and avoid causing harm, transparency in the characteristics of data sets used in AI models to define their limitations and appropriate patient cohorts, and protection of patient confidentiality.

“We are at a truly transformational moment in oncology with the emergence of generative AI and the development of LLMs,” Dr Haddad said. “This is anticipated to have an unprecedented impact on cancer research and care delivery.”  

Disclosures: Dr Elemento is the co-founder of OneThree Bio, a company that uses AI to discover and develop new drugs in the oncology space. Dr Margolies serves on the medical advisory board for Screenpoint Medical. Dr Haddad reported having no relevant disclosures.

References

1. Shreve JT, Khanani SA, Haddad TC. Artificial intelligence in oncology: Current capabilities, future opportunities, and ethical considerations. Am Soc Clin Oncol Educ Book. 2022;42:842-851. doi:10.1200/EDBK_350652

2. Artificial intelligence and machine learning (AI/ML)-enabled medical devices. US Food and Drug Administration. Published October 5, 2022. Accessed September 24, 2023.

3. Luchini, C., Pea, A. & Scarpa, A. Artificial intelligence in oncology: Current applications and future perspectives. Br J Cancer. 2022;126:4-9. doi:10.1038/s41416-021-01633-1

4. Schaffter T, Buist DSM, Lee CI, et al. Evaluation of combined artificial intelligence and radiologist assessment to interpret screening mammograms. JAMA Netw Open. 2020;3(3):e200265. doi:10.1001/jamanetworkopen.2020.0265

5. Arasu VA, Habel LA, Achacoso NS, et al. Comparison of mammography AI algorithms with a clinical risk model for 5-year breast cancer risk prediction: An observational study. Radiology. 2023;307:5. https://doi.org/10.1148/radiol.222733

6. Shen Y, Shamout FE, Oliver JR, et al. Artificial intelligence system reduces false-positive findings in the interpretation of breast ultrasound exams. Nat Commun. 2021;12(1):5645. doi:10.1038/s41467-021-26023-2

7. Mansur A, Saleem Z, Elhakim T, Daye D. Role of artificial intelligence in risk prediction, prognostication, and therapy response assessment in colorectal cancer: Current state and future directions. Front Oncol. 2023;13:1065402. doi:10.3389/fonc.2023.1065402

8. FDA authorizes marketing of first device that uses artificial intelligence to help detect potential signs of colon cancer. US Food and Drug Administration. Published April 9, 2021. Accessed September 24, 2023.

9. Senthil Kumar K, Miskovic V, Blasiak A, et al. Artificial intelligence in clinical oncology: From data to digital pathology and treatment. Am Soc Clin Oncol Educ Book. 2023;43:e390084. doi:10.1200/EDBK_390084

10. You Y, Lai X, Pan Y, et al. Artificial intelligence in cancer target identification and drug discovery. Signal Transduct Target Ther. 2022;7(1):156. doi:10.1038/s41392-022-00994-0

11. Wang L, Song Y, Wang H, et al. Advances of artificial intelligence in anti-cancer drug design: A review of the past decade. Pharmaceuticals. 2023;16(2):253. doi:10.3390/ph16020253

12. Hindocha S, Zucker K, Jena R, et al. Artificial intelligence for radiotherapy auto-contouring: Current use, perceptions of and barriers to implementation. Clinical Oncology. 2023;35(4):P219-226. doi:10.1016/j.clon.2023.01.014

13. Qureshi TA, Javed S, Sarmadi T, Pandol SJ, Li D. Artificial intelligence and imaging for risk prediction of pancreatic cancer: A narrative review. Chin Clin Oncol. 2022;11(1):1. doi:10.21037/cco-21-117

14. Haver HL, Ambinder EB, Bahl M, Oluyemi ET, Jeudy J, Yi PH. Appropriateness of breast cancer prevention and screening recommendations provided by ChatGPT. Radiology. 2023;307(4):e230424. doi:10.1148/radiol.230424

15. Pan A, Musheyev D, Bockelman D, Loeb S, Kabarriti AE. Assessment of artificial intelligence chatbot responses to top searched queries about cancer. JAMA Oncol. Published online August 24, 2023. doi:10.1001/jamaoncol.2023.2947

16. Chen S, Kann BH, Foote MB, et al. Use of artificial intelligence chatbots for cancer treatment information. JAMA Oncol. Published online August 24, 2023. doi:10.1001/jamaoncol.2023.2954

17. Nunez J-J, Leung B, Ho C, Bates AT, Ng RT. Predicting the survival of patients with cancer from their initial oncology consultation document using natural language processing. JAMA Netw Open. Published online February 27, 2023. doi:10.1001/jamanetworkopen.2023.0813

This article originally appeared on Cancer Therapy Advisor