How digital twins are making cancer surgery safer

Surgical procedures are a common part of cancer diagnosis and treatment, from a biopsy to determine the stage and type of cancer to the removal of a cancerous tumor from the body. Surgery continues to evolve as new advances and even small improvements make procedures easier for both surgeons and patients.

Jeff Siewerdsen, Ph.D., and his team are using data science and virtual models, like digital twin technology, to create these advances and to test new surgical technology and workflows before using them on patients.

“We are working to bring data-intensive models and digital twins to improve safety and quality in cancer interventions,” says Siewerdsen, professor of Imaging Physics and co-lead of Safety, Quality and Access within the Institute for Data Science in Oncology (IDSO). “These models are helping to accelerate the translation of valuable, cutting-edge technologies to clinical impact.”

The power of models

IDSO and its focus area on Safety, Quality and Access is using virtual modeling to improve outcomes for patients. One type, called surgical process modeling, can help illustrate — in simpler terms — complex surgical workflows and help to determine potential benefits of emerging technologies.

“There’s an old saying: ‘No model is true, but some models are useful,’ and surgical process models are definitely in that ‘useful’ category,” Siewerdsen says. “A model gives us the power to ask, ‘what if?’ and to make a prediction.” The models can also provide answers in a quicker, more efficient and sometimes more ethical way than could be achieved with real-world experimentation.

Medical imaging, for example, uses models to help determine how to design a new imaging system and what amount of time or radiation dose is required to create an acceptable image for diagnosis. It would be far more difficult and expensive to create a machine, realize some shortfall in clinical trials, and then start from the beginning to design a new one. Instead, models of patient anatomy and models of the imaging system provide evidence of what the image quality would be under different conditions.

Models are important in medical treatment as well. There is an entire field of “model-guided therapy,” which uses digital twins and virtual clinical trials to test treatments on models before doing so in real patients, potentially identifying the most effective treatment for each individual patient. Virtual clinical trials are an emerging area that is recognized in the scientific community and can also help meet regulatory requirements by showing that a new technology is at least as good as the existing options.

“A virtual clinical trial runs on a model,” Siewerdsen says, “and surgical process modeling could provide evidence of the value of an emerging technology for a given surgical task.”

Robotic lung biopsies

One clinical example is lung nodule biopsy via bronchoscopy. Siewerdsen’s work with collaborators like Roberto Casal, M.D., shows that cone-beam computed tomography (CBCT) with robotic-assisted bronchoscopy is better than conventional approaches of 2D fluoroscopy and ultrasound. Although it takes longer and can expose the patient to slightly higher levels of radiation, both stay at clinically acceptable levels, and CBCT-guided robotic-assisted bronchoscopy can have diagnostic success of above 90%, compared with around 60–70% with conventional methods.

“The improvement in geometric error leads to more accurate biopsies and improved diagnostic yield,” Siewerdsen says. “That is now being borne out in clinical studies.”

Surgical process modeling can also help explain the sources of such an improvement. For example, they can show how much of the increase in diagnostic yield is attributable to 3D cone-beam computed tomography and how much is from the robot. The gains in geometric accuracy might come from the 3D CBCT; whereas, the robot might provide a more streamlined workflow for approaching and navigating to the lesion and providing a steady platform during biopsy sampling. This parsing of effects could help with practical decisions, such as helping a biopsy clinic determine which of the two to buy if the whole system was prohibitively expensive.

“The model provides quantitative evidence for the major advantages of 3D imaging for transbronchial biopsy, and those systems are now emerging in clinical endoscopy suites, where previously there was just 2D fluoroscopy and ultrasound,” Siewerdsen says. “And collaborators at UT MD Anderson were the true trailblazers in bringing this valuable technology to clinical use.”

Image‑guided spine surgery

Siewerdsen and his team have also examined image-guided spine surgeries using surgical process modeling, which include spine surgery target localization with long-length imaging and pedicle screw placement with intraoperative CT and navigation. His models show that new technologies for this procedure decrease procedure time and improve accuracy.

Just as in 3D bronchoscopy, the model helps quantify the advantages of 3D navigation, and although early adopters have been using those systems for many years, the technology is only prevalent in about 20% of spine surgery practice.

“The model highlights the workflow issues that are part of the reason for limited adoption and gives insight on why surgeons may feel that navigation slows down the case, even though it doesn’t take more of the surgeon’s time, especially for long spinal constructs,” Siewerdsen says. “It also shows the need for navigation systems to maintain accurate registration throughout the case, even in the presence of complex anatomical motion.”

The model has, in some ways, already affected real-life technologies being used in the operating room. For example, the image‑guided spine model demonstrates the clinical value of long-length imaging systems that are now emerging. The next steps for Siewerdsen and his team will be modeling other imaging systems with semiautonomous motion and surgical robotics, both of which have immediate real-world impact.

Surgical process modeling is also integral to new collaborations between UT MD Anderson and researchers at The University of Texas at Austin, including the IG-RABIT project that is creating new technologies for image-guided orthopedic cancer surgery. "In this project, surgical process models are being used to evaluate the benefit and optimize the workflow of automatic surgical planning, patient-specific implants and surgical robotics," Siewerdsen says.

What’s next for using digital twins to improve surgery

Implementation of surgical process modeling can be extended not only to new procedures — including surgery and interventional radiology — but also to model specific, individual surgeries. Such a patient-specific process model takes in all the information about the individual patient, the operating team and the technologies available and runs thousands of simulations to predict how the case will go, what variations will improve the surgery, which steps may be most prone to challenge or error, and more.

“That gets at the digital twin concept and is very exciting,” Siewerdsen says. “Extending surgical process models to patient-specific digital twins on a regular basis will be a major effort. But how cool that would be: a new approach to optimizing the surgical plan and improving patient outcomes.”

Learn more about the Institute for Data Science in Oncology (IDSO) at UT MD Anderson.