The use of AI technology has transformed the field of Radiation Oncology treatment planning. This is particularly true with the introduction of AI auto-contouring, which promises to improve the planning workflow by decreasing the time spent on OAR contour creation and increasing overall plan quality.
AI auto-contouring has sparked an ongoing discussion among both veteran and new adopters as they evaluate whether this technology is indeed improving clinical practice. As many would agree, the implementation of an AI auto-contouring solution should occur without disruption to the department’s treatment planning quality and without additional effort to an already complex process.
As you embrace the future of AI auto-contouring in your clinic, consider how the following implementation factors may be impacting your ability to provide accurate and efficient treatment planning.
If your software does not start the AI auto-contouring process automatically—without clicks or wait times—is it really automatic? Auto-contouring should begin as soon as your patient is simulated. For true zero-click segmentation, contours need to be readily available in the system when you need them. Manual launch costs time and creates unnecessary steps in the overall workflow.
When implementing new technology like AI auto-contouring, it is important to consider the pathway that provides the most efficiency and accuracy with the least amount of workflow disruption. A solution that does not add steps or pull you away from the contouring system you prefer to work in is essential to streamlining the process. If your chosen AI auto-contouring solution limits your ability to customize your work environment, is it really working for you?
Is your AI auto-contouring solution capable of both local and cloud deployment? Depending on your department’s needs, you may need one, the other, or both. Sometimes the answer isn’t clear until you begin planning and testing the implementation. Adopting a solution that offers deployment flexibility is especially important.
Even relatively good AI auto-contour models may not always produce accurate results, which can lead to errors in treatment planning. Accuracy is paramount when it comes to AI auto-contouring in order to improve efficiency and enhance the quality of treatment planning. If your adopted AI auto-contouring solution produces inferior results, it will likely never be trusted by experienced dosimetrists.
Most hospital systems and individual clinics have existing sets of unique contour names, colors, and codes in use. If your AI auto-contouring solution does not adhere to your department’s established standards of identification, you will be required to manually edit each set of contours for every case. This can be a tedious and time-draining process.
Although AI auto-contouring can speed up the segmentation process, the results still require review—and sometimes input—from the dosimetrist. Clinicians may be burdened with an increased workload if significant errors occur and inadequate tools are available to correct the contours. Inadequate correction tools add time and uncertainty to the contour QA process.
AI auto-contouring solutions with large datasets may seem desirable, but these contours can have limited applicability in clinical practice. AI models based on large, generalized datasets will lack specific and accurate case examples. Datasets used specifically for training AI models should accurately and consistently follow a set of widely accepted guidelines in order to provide good results. When considering AI vendors, focus on whether the datasets are relevant to your needs. Is your vendor focusing on a variety of data and accuracy for the application at hand, or focused only on having a high number of datasets?
There is growing evidence that the adoption of AI technology across the healthcare sector will result in significant cost savings, as much as $200 billion to $360 billion per year. For radiation oncologists, the careful selection of an AI auto-contouring solution that increases productivity, reduces errors, and enhances the dosimetrist’s workflow is of the utmost importance for success and cost savings.
As radiation oncologists continue to embrace the future of AI in clinical practice, it is important to consider how these and other implementation factors may affect the accuracy and efficiency of treatment planning. Implementing an AI auto-contouring solution in your clinic should never compromise the quality of treatment planning or add additional complexity to your process. The right solution ensures that AI auto-contouring is integrated effectively and responsibly into your clinic.
Want to access more information about the power of flexible implementation for AI auto-contouring? Click here.
Jay Obman is a Product Manager at MIM Software. Jay works closely with clinics across the country to create impactful auto-contouring solutions that address clinical workflow bottlenecks and lead the way in flexible implementation.