Cancer: How AI architects are constructing the future of radiotherapy

Radiotherapy¹ to treat cancer is a complex process, involving a number of different steps and technologies. Planning radiation treatment takes a long time – currently about two weeks, from when therapy is prescribed until treatment actually starts. "When it comes to cancer, time is a crucial factor. It can make the difference between life and death in some cases," Fernando Vega notes earnestly.

Vega sits, focused on this computer, with multiple monitors displaying CT, PET-CT and MRI scans of a cancer patient. The tumor growing in the left half of a patient's brain shows brightly visible and distinctly contrasted on the screens.

Fernando Vega, who originally hails from Colombia, is a computer scientist and specialist in medical imaging and software applications. At Siemens Healthineers, he leads Software & Concept Definition at the “Cancer Therapy Imaging" unit of the Varian Business Area. "One question we ask ourselves is: How can we optimize and speed up the entire radiation therapy planning and delivery workflow – with the help of AI?"

This is the question that Vega and a large team of colleagues from various business areas are jointly seeking to answer. While Siemens Healthineers already offers a comprehensive portfolio of equipment and systems for radiotherapy, AI integration is poised to soon make much more possible.

In this first step, the radiation oncologist² examines the findings to date and discusses the potential benefits, risks and type of radiation with the cancer patient.

The radiation oncologist enters this information into an oncology information system. The physician's scheduling staff then coordinate with the patient to set up an imaging appointment, as highly precise medical images provide the basis for subsequent treatment. Ideally, an appointment is scheduled for the patient within a few days. This completes the first step in the process. 

Vega then outlines how the scenario would play out in the workflow improved by AI. The initial substeps of consulting with the patient, prescribing radiation therapy, and data input in the information system would remain the same. However, the follow-up appointment scheduling would be controlled by AI in the oncology information system. Vega explains: "The AI would already know exactly what type of scan the patient needs based on the stored information, and could also see which suitable scanner is available the soonest. This could reduce waiting times."

Part of the planning process consists of what's called contouring – a complex and time-consuming step. On the generated medical images (i.e. the CT, PET-CT and/or MRI scans), the area to be treated – such as the tumors – are delineated from the surrounding healthy tissue. The organs-at-risk  (OARs, for short) which need to be protected during the radiation treatment are clearly identified. 

Vega explains that, in standard clinical practice depending on the given country or hospital, this contouring of the OARs is performed by resident physicians of the radiation oncology department, medical technical radiology assistants (MTRA), medical physicists³, or medical dosimetrists. However, this manual process is susceptible to error. Automatic contouring could already lead to improvements today, providing automated contouring of the OARs.

The medical physicist – and thus the patient as well – is meanwhile waiting. Only once the OARs and target areas have been clearly defined and approved can the medical physicist begin calculating the exact dose of the individual radiation sessions in what is termed the fractionation. 

This dosimetry plan, too, must in turn then receive the "blessing" of the radiation oncologist. "What's more, these physicians are generally very busy in their daily clinical routines, and often difficult to reach," Vega adds. Yet, this step must also be completed before the patient can finally be brought in to begin radiation treatment.

Based on the results of precise automated contouring, planning software could automatically calculate the correct dose plan for radiotherapy. The AI-controlled oncology information system that supports all steps across the entire process could notify the radiation oncologist automatically that the plans are ready for final review. 

The team in charge of planning the radiation treatment would be informed in real time by the information system of which radiation planning has been completed. Considering the available systems and personnel, it could assign each patient the earliest possible appointment to begin treatment. In the overall process of preparing radiotherapy plans, this would reduce not only the "ping-ponging" of activities and thus the waiting times, but also any potential sources of error.

Vega then explains today's current workflow for the fourth step in the whole process: the actual radiation treatment. The tumor is treated with X-rays using a linear accelerator (LINAC). The objective is to kill the cancer cells and prevent further cell division from occurring. The treatment appointments at which dose fractions are administered are generally scheduled over multiple weeks.

How could AI improve the workflow of the actual radiation treatment? "In cases of lung cancer, for instance, the tumors could be identified automatically in the image and the radiation predictively adjusted in real time to the patient's breathing movements. This could give us even more precise results," Vega explains.

Summing it all up: Radiotherapy planning today is marked by significant loss of time because the medical team has to coordinate various systems and workflows. The process takes on average several weeks.

Vega says that if all of these systems could in future think "as one" thanks to AI, there would be practically no unnecessary waiting times: "Our vision is to shorten the wait to a just few hours between when radiotherapy is prescribed and actual radiation treatment begins."

What's more, AI would enable all therapy to be custom-tailored more individually to each patient, leading to better treatment results – and thus, combined with the time savings, helping people live longer. 

It would also benefit daily clinical routines, as AI can accelerate repetitive, recurring tasks. This can ease the burden on clinical staff and help cut costs by enhancing efficiency.

One thing is clear: an undertaking like this, that comprises so many work steps and different devices and systems, cannot be done by one person. As Fernando Vega – who very soon needs to join a video call with his international colleagues – knows well, it takes many keen minds.

"We still have a long way to go until we will be able to implement the full vision on our devices and systems." Vega is well aware of this fact – especially considering that the predevelopment project would need to be scalable. Moving forward, additional areas such as imaging for early cancer detection and laboratory diagnostics could be integrated into the AI-optimized workflow. The objective of all this is to establish a solid, networked solution for cancer centers that enable comprehensive, full-service care along the entire cancer treatment pathway – from early detection to follow-up care.

In any case, the inveterate innovator Fernando Vega, who even listens to audio books about AI while jogging after work, is fascinated by artificial intelligence and the potential it holds for the future. Yet, all the while remaining circumspect:

© Photos: Markus Ulbrich
© Video: Lisa Fiedler (camera, editing); Markus Ulbrich (camera); Cagdas Cubuk (camera, sound); Katja Gäbelein (concept, direction) 
© Motion Graphics: Viola Wolfermann