Researchers Leverage Generative AI to Improve Cancer Treatment Targets

Researchers from the University of Texas at San Antonio (UTSA), UT Health San Antonio, and the University of Pittsburgh have developed a generative artificial intelligence (AI) tool to improve adaptive radiotherapy, according to a study published in the May 2023 edition of Medical Image Analysis. 

Adaptive radiotherapy (ART) is an advanced type of cancer therapy in which treatment doses are continually adjusted based on changes in patient anatomy, such as weight loss or tumor shrinkage, or at pre-defined intervals to administer the most accurate dose of radiation possible. 

In doing so, clinicians have the potential to target tumors more effectively, reduce the amount of healthy tissue exposed to radiation, and minimize some side effects of treatment.  

However, the study indicates that cancer patients undergoing radiotherapy are given a computed tomography (CT) scan to allow clinicians to identify where in the body a tumor is located. This CT image, and the location of the tumor, are then used to make a treatment plan to remove the tumor via targeted radiation doses. 

Often, cone-beam computed tomography (CBCT) is leveraged over the course of a treatment plan, typically weekly or after each dose, to determine how much a tumor has shrunk and adjust dosage levels to ensure as little healthy tissue is damaged as possible. 

However, CBCTs are low-quality images that can be difficult to read and easy to misinterpret. To address this, the researchers turned to AI. 

“This is a multidisciplinary research project that includes multiple faculty members who have come together with a different skillset – AI, data analytics, and health care – to solve a challenge,” said Paul Rad, PhD, UTSA associate professor of information systems and cyber security, in a press release discussing the study’s findings. 

“Our study aimed to analyze treatment doses administered and develop a precise map of a patient's cancer progression while accounting for potential variability using uncertainty estimation," he explained. 

The research team leveraged domain adaptation techniques to pull high-quality textural and spatial features from pre-treatment CT scans and tumor shrinkage information from weekly CBCTs. The AI then helps visualize the tumor region affected by a patient’s radiotherapy, allowing for more accurate tumor evaluation. 

The study cohort consisted of 16 cancer patients whose pre-treatment CT and mid-treatment weekly CBCTs were captured over the course of six weeks. 

Using the AI tool, the researchers demonstrated improved tumor shrinkage predictions and a 35 percent decrease in the risk of radiation-induced pneumonitis, a type of lung injury that can occur following radiotherapy. 

This approach could help clinicians more reliably track how much a tumor has decreased each week and enable them to plan the upcoming weeks’ dosages with more precision. Overall, this higher level of precision has the potential to significantly improve tumor targeting and better personalize ART planning, the researchers noted. 

However, more research is needed before an AI-assisted ART approach can be implemented in clinical settings.  

“Besides just building more advanced AI models for radiotherapy, we also are super interested in the limitations of these models,” said Arkajyoti Roy, PhD, UTSA assistant professor of management science and statistics. “All models make errors and for something like cancer treatment it’s very important not only to understand the errors but to try to figure out how we can limit their impact; that’s really the goal from my perspective of this project.”