Artificial Intelligence, Machine Learning Are Poised to Transform Radiation Oncology

Imaging

AI applications for diagnostic image analysis led the way in radiation oncology; tumor staging and treatment planning applications for medical imaging data have already been developed and FDA-approved, and more of these tools are in development.^4-7

Predictive AI can inform and improve treatment planning and prognostication, for example, by delineating the range of tumor motion during radiation delivery to better optimize dose delivery to target organs while sparing healthy, nontarget tissues, or by identifying patients at risk of radiotoxicities such as radiation fibrosis in irradiated neck muscles or radiation pneumonitis in patients undergoing radiotherapy for non-small cell lung cancer.⁸ AI can compare sequential patient images to rapidly and accurately detect disease progression or treatment response, techniques that can improve adaptive radiotherapy plan adjustments. For example, Varian Oncology Systems (Palo Alto, California) developed the ETHOS^TM AI system for streamlining adaptive radiotherapy.⁵

AI imaging-related applications can help perform at least 6 broad types of image data-based or related tasks in radiotherapy^4-9:

• Automated cancer detection AI can be used to quickly and accurately detect tumors, streamlining and hastening diagnostic imaging interpretation.
• Segmentation AI algorithms can automatically identify and delineate or segment anatomic features in medical images, and identify and delineate (contour) volumes of interest.
• Classification AI algorithms can stage and potentially identify tumor subtypes.
• Treatment planning AI can contour volumes of interest, such as gross tumor volume and treatment volume, and can propose optimal external-beam radiotherapy beam arrangements and patient positioning strategies. It can also assist with dosimetry and predictive dose calculations, treatment simulation modeling, and image-guided adaptive radiotherapy, adjusting proposed beam arrangements or patient positioning to better target tumors that are growing or shrinking between radiotherapy doses.⁶  
• Treatment set-up AI can help ensure consistent patient positioning during treatment set-up, automatically tracking patient position and hastening treatment preparation during radiotherapy appointments.
• Monitoring AI can monitor for changes indicating tumor progression or treatment response using sequential images acquired over time, from diagnosis or the time treatment is initiated. During treatment, this can involve image-guided adaptive radiotherapy (mentioned above) and assessments of treatment effectiveness.

Patient Management

One goal of AI tools in the radiation oncology clinic is to predict radiation-associated toxicities and patient adverse events, rather than reacting to them after they emerge.⁸AI is well positioned to identify and consider larger numbers of patient comorbidities and risk factors for these problems, and to incorporate treatment plans and anatomic image features to more accurately predict management challenges such as oral mucositis, dysphagia, or xerostomia, and offer guidance so that the care teams can anticipate and manage them.⁸ AI tools have been developed for predicting pneumonitis, esophagitis and brain tumor-associated seizures.^8,10

AI’s ‘Black Box’

Although some AI tools have been FDA-approved, most remain investigational and under development.⁸ One major challenge is the explainability of AI-generated decisions, AI’s so-called “black box” problem: human users frequently cannot readily understand how AI tools arrived at a particular conclusion or recommendation, complicating validation and consideration of alternative solutions to a given problem.⁸

Take radiomics, for example. Radiomics involves computer extraction of features from patients’ image data that are associated with tumorigenesis, tumor presence or growth that might elude human radiologists in visual image examinations, based on system “training” with massive sets of patient CT scans.¹¹ That has obvious clinical utility. Machine learning systems are “trained” on large sets of medical images to accurately detect, demarcate, and even stage tumors — all using features that are not visible to the human eye and are not explained by the system to its human users. AI systems cannot always explain the features informing a particular conclusion.

For example, it was recently reported that a machine learning system developed at MIT can predict short-term and long-term lung cancer risk — up to 6 years before diagnosis — based on data from a single low-dose CT scan.¹² The system learns to identify high-risk image data but it is not clear what it is learning: the specific image data features that are leading to prediction of lung tumors. Does it detect early microvascular or tissue texture changes at microtumors? It’s not known.

References

1. Wahid KA, Glerean E, Sahlsten J, et al. Artificial intelligence for radiation oncology applications using public datasets. Semin Radiat Oncol. 2022;32(4):400-414. doi:10.1016/j.semradonc.2022.06.009
2. Massat MB. The intersection of radiomics, artificial intelligence and radiation therapy. Appl Rad Oncol. 2019;8(4):34-37.
3. AFDO/RAPS Healthcare Products Collaborative. Bias in artificial intelligence in healthcare deliverables. Accessed February 7, 2023. https://www.healthcareproducts.org/wp-content/uploads/2022/10/Final-v2-Bias-in-Artificial-Intelligence-10.27.22.pdf
4. Furlow B. Deep learning poised to revolutionise diagnostic imaging. Lancet Respir Med. 2017;5(10):779. doi:10.1016/S2213-2600(17)30292-8
5. Chamunyonga C, Edwards C, Caldwell P, Rutledge P, Burbery J. The impact of artificial intelligence and machine learning in radiation therapy: considerations for future curriculum enhancement. J Med Imaging Radiat Sci. 2020;51(2):214-220. doi:10.1016/j.jmir.2020.01.008
6. Riegel AC. Applications of artificial intelligence in head and neck radiation therapy. Appl Rad Oncol. 2021;10(1):6-12.
7. Luchini C, Pea A, Scarpa A. Artificial intelligence in oncology: current applications and future perspectives. Br J Cancer. 2022;126(1):4-9. doi:10.1038/s41416-021-01633-1
8. Huynh E, Hosny A, Guthier C, et al. Artificial intelligence in radiation oncology. Nat Rev Clin Oncol. 2020;17(12):771-781. doi:10.1038/s41571-020-0417-8
9. Deig CR, Kanwar A, Thompson RF. Artificial intelligence in radiation oncology. Hematol Oncol Clin North Am. 2019;33(6):1095-1104. doi:10.1016/j.hoc.2019.08.003
10. Liu Z, Wang Y, Liu X, et al. Radiomics analysis allows for precise prediction of epilepsy in patients with low-grade gliomas. Neuroimage Clin. 2018;19:271-278. doi:10.1016/j.nicl.2018.04.024
11. Kocher M. Artificial intelligence and radiomics for radiation oncology. Strahlenther Onkol. 2020;196(10):847. doi:10.1007/s00066-020-01676-y
12. Mikhael PG, Wohlwend J, Yala A, et al. Sybil: a validated deep learning model to predict future lung cancer risk from a single low-dose chest computed tomography.  J Clin Oncol. Published online January 12, 2023. doi:10.1200/JCO.22.01345