Artificial Intelligence (AI) has the potential to transform radiation oncology and improve the quality of care for patients with cancer. In last month’s blog, the fundamentals of AI were covered and the implementations for CT simulation, treatment planning, quality assurance (QA), and treatment delivery were discussed. Those applications included enhancing low-quality scans, predicting abnormal sensitivities and anatomy, auto-contouring, auto-planning, QA assistance, ensuring correct delivery to the patient, and re-planning intervention.
Based on predictions, approximately 20% of clinical practices may be using Deep Learning (DL) based tools clinically within the next few years. Therefore, it is valuable to understand how this technology will affect the day-to-day workflow for technical professionals in radiation oncology. Additionally, the barriers to the implementation of AI in radiotherapy will be discussed.
The availability of AI tools will certainly change the composition and skillset of the radiation oncology workforce, especially those that currently spend substantial amounts of time on repetitive tasks requiring manual input. AI will predominantly affect staff members that perform the technical aspects of radiation with less of an effect on activities involving direct interaction with patients. The technical team of a radiation oncology center includes Radiation Oncologists, Physicists, Dosimetrists, and Radiation Therapists. Although there is a multitude of other professionals working in a cancer center, such as nurses, they do not work as directly with treatment planning and delivery; therefore, they will not be discussed in this section.
Radiation Oncologists perform segmentation tasks similar to a dosimetrist, this includes contouring organs at risk (OARs) and tumor volumes. As AI-based segmentation algorithms begin to replace manual segmentation, the focus of the physicians will shift towards “front-of-house” activities, such as patient counseling, education, support, and clinical management. Furthermore, academic teaching and research will increase at cancer centers due to the reduction of time spent on repetitive tasks.
On average medical physicists spent most of their time performing clinical service and consultation. AI applications have the potential to reduce the frequency and time of routine QA tasks; thus, causing a shift in the focus of medical physicists towards high-risk problems and the development and implementation of new technologies. Furthermore, the shift will likely cause the transition of medical physicists into more direct patient role responsibilities. Despite these changes in the role of a medical physicist, they will continue to be instrumental in ensuring the accuracy and precision of clinical treatment.
Segmentation, contouring, and manual treatment-planning tasks are going to be largely impacted by AI. Plans developed by AI processes are comparable to those manually generated by dosimetrists while greatly reducing the planning time. In the short term, we expect that dosimetrists will focus on more high-risk and complex situations that present a challenge for current machine learning approaches. Additionally, a reduction in time spent on contouring and planning may lead to an increase in research and special procedures. Currently, there are widespread understaffing issues in radiation oncology centers and the automation of the repetitive tasks will greatly aid in the quality and efficiency of patient treatment. Although AI integration will help dosimetrists in the short term, full automation of the treatment-planning process might lead to a reduction in the number of dosimetrists in the future.
AI has multiple applications for ensuring the correct delivery of treatment to patients. These software tools will help radiation therapists ensure accurate and safe treatment, as well as increase the overall productivity of radiotherapy centers. Since radiation therapists have the most direct patient contact of the technical team, they will continue to have an important role in centers and are unlikely to experience a demand reduction.
Figure 1- Role Shifts in Radiation Oncology. Figure from Huynh, Elizabeth, et al. “Artificial Intelligence in Radiation Oncology.” Nature News, Nature Publishing Group
Barriers to Implementation
There are multiple barriers to the adoption of machine-learning processes in radiation oncology. Firstly, AI requires not only a monetary investment but also a time investment in order to understand the uses and limitations of these tools. High dependence on new technology without the proper training can result in misuse and result in harm to the patient. Proper training is further hindered by the closed-source software of the DL algorithms used in the machinery.
Secondly, most AI tools remain at the prototype stage and lack external review, this issue can be contributed to the proprietary software of machine learning algorithms by the vendor. Establishing trust in AI systems is essential for adoption into a clinical setting.
Lastly, the field of radiation oncology is highly subjective. Treatment plans vary depending on the preferences of each radiation oncologist; therefore, it is important to note the difference between an optimal plan and a preferred one. This issue is further exacerbated by the lack of standardization of treatment plans. The fundamental process of machine learning is based on training data and the lack of uniformity in treatment plans greatly reduces the capability of the algorithms. In other words, machines can be no more “correct” than the human input given. Ultimately, the fundamental challenge in AI implementation lies in the standardization of existing clinical practices.
Conclusion and Further Resources
Implementation of AI solutions has the potential to transform radiation oncology and the overall quality of radiation therapy for patients with cancer. These improvements will result in significant changes for the radiation oncology workforce. Despite the long-term job security threats of AI adoption for technical professionals in radiotherapy, centers are currently understaffed and under-resourced so the demand for these services is still increasing. Overall, all members of the technical team are and will be vital for the quality, safety, and efficiency of radiotherapy centers for the foreseeable future.
For more information on the process and applications of AI in radiation therapy read the Artificial Intelligence and Radiotherapy Part 1 Blog. Further resources are included in the links below.