The Use of AI in Dosimetry
By: Tammy McClausland
This blog post is adapted from “Is AI Reshaping the Medical Dosimetrist’s Role,” which was published in Radiation Oncology News for Administrators, Vol. 32 No. 2.
A 2020 Nature article says, “AI could have particularly transformative applications in radiation oncology given the multifaceted and highly technical nature of this field of medicine with a heavy reliance on digital data processing and computer software. . . . AI has the potential to improve the accuracy, precision, efficiency and overall quality of radiation therapy for patients with cancer.”1 Two medical dosimetrists share how they’re using AI and its implications for their profession and patient care.
AI in Medical Dosimetry
Innovations on the AI front, Brian Napolitano says, “are hopefully going to allow dosimetrists to work smarter and to automate some of the more routine tasks what we do so that dosimetrists have the opportunity to practice at the top of their skillset.” Over his career, Napolitano, the director of Medical Dosimetry at Mass General Cancer Center in Boston, has seen lots of technological innovations, CT-based planning, intensity-modulated radiation therapy (IMRT), volumetric modulated arc therapy and proton therapy.
AI technology is “not nearly as sophisticated as you would imagine it would be. Many of the AI applications we use are specific to making our workflows more proficient, saving time upfront in order to avoid creating plans from scratch,” says Rihan Davis, chief medical dosimetrist at the Robert Wood Johnson University Hospital Radiation Oncology Department in New Jersey.
Radiation therapy plans generally take one week to generate, but the processes can be condensed using AI. An auto contouring AI program enables medical dosimetrists to delineate organs at risk [OARs] within an hour rather than half a day. “Now we can focus on the actual planning and fine tuning the plan in delivering more dose to the tumor while sparing the OARs. Patients live longer, and they have better control of the disease while sustaining a lot less toxicity,” says Davis.
Napolitano explains, “Some clinics have started to look at incorporating automated plan development pieces, which is exciting because it ensures that dosimetrists have access to a lot of tools that help them seek out the best possible quality treatment plan. AI can ensure a level of quality, efficiency and safety for our patients.”
The Human Element Remains Critical
AI automation is intended to compliment the medical dosimetrists’ skill sets, says Napolitano, but the human element is required to evaluate what the AI does and to troubleshoot and course correct as needed. “There’s certainly a concern, not necessarily specific to dosimetrists but more widespread, that the incorporation of AI may allow people to do things in a more streamlined fashion, but that they may not realize when an AI process may have a hiccup or be slightly broken or off kilter,” he says.
“No matter how sophisticated the AI has become, it’s still a machine at the end of the day,” says Davis. “A human evaluator is always required in order to evaluate the final product. I don’t foresee AI eliminating the dosimetrist’s role anytime in the near future.”
AI also needs to be trained. For example, Davis says today we might evaluate a V20—a Vx, where x is a numeric value—in order to associate toxicity to the 20 Gray isodose line. Newly published research may now require us to evaluate a V10 or a V5. “With such instances, AI would have to be retrained. It would require the development of new models and retraining the AI processes,” she says.
At Davis’ institution, to demonstrate AI’s value, they compare the AI-generated plan to some of their previous plans. “AI expedites the proficiency in my planning, but 90 percent is still
being tweaked manually just to fine tune the plan. Ultimately AI will always create a more efficient process, but I don’t think, we would get to a scenario where AI is performing all the tasks of a dosimetrist,” she says.
AI in the Workplace and in Student Training
The medical dosimetrist’s role will continue to evolve. “When IMRT was introduced, people thought the incorporation of cost-based function inverse planning was going to reduce the number of dosimetrists that were necessary. In fact, more dosimetrists were needed because the technology was increasingly more complex,” says Napolitano.
As the incidence of cancer continues to increase, so will the number of patients treated with radiation. “AI will help to alleviate some of the burdens on the demand for dosimetrist skills and allow them to focus their efforts where they can be most valuable, where their skills and their subject matter expertise is most needed,” he says.
At Davis’ institution, medical dosimetry students receive one year of clinical training. Even if AI is used for treatment planning or contouring, students perform the duties manually during their education. Without such training, she says, students wouldn’t be able to identify if some of the OARs were in trouble, for example. If a student doesn’t know how to contour the kidney and understand the proximity to the disease and recognize what the safe dose is that can be delivered, they wouldn’t be able to assess whether AI made a mistake contouring that organ. “That’s why the profession’s governing bodies are pretty conservative and rigorous in what we test students on. We are required to have them run through those exercises to perform those duties,” says Davis.
AI will inevitably change medical dosimetry training both with regard to curricula and in the clinic. “It’s only a matter of time before dosimetrists who are in training will have a level of understanding and appreciation for the guts of how the AI is arriving at the solution to whatever it might be tasked with,” says Napolitano.
Future Use of AI
Administrators should keep in mind that AI tools are evolving rapidly. Vendors are putting their development efforts into enhancing these tools. He says dosimetrists, physicists and physicians can be part of the dialogue about how best to use AI tools, how best to integrate them and how best to support their use.
“AI is going to be a bigger part of how we treat patients with radiation. The more AI tools are refined and enhanced, the more patients will benefit,” he says. “Administrators should recognize that AI is a complement to, not a replacement for, dosimetrists. AI’s intended to be something that complements dosimetrists’ skill sets and integrates their skills into how we optimize quality, safety and efficiency for patient care.”
References:
Huynh E., Hosny A., Guthier C., et al. Artificial intelligence in radiation oncology. Nature Reviews Clinical Oncology. 17: 771–781.
How is your cancer center using AI in treatment planning?
Share any helpful tips and suggestions.
We would love to hear your experiences.
Share your thoughts here, or login to SROA Connect and join the conversation. If you are not a member of SROA yet, learn more about joining the radiation oncology association serving the niche profession of Radiation Oncology Administrator.
Related Content:
Radiation Oncology News for Administrators, Vol. 32 No. 2
SROA Blogs
2020 Nature article
Links:
Society For Radiation Oncology Administrators (SROA)
Comments