Study demonstrates how AI can develop more personalized cancer treatment strategies

This study shows that combining the power of mathematical modelling and AI can have a profound impact on clinical cancer care, improving effectiveness and reducing costs.

Professor Philip Mainiis director of the Wolfson Centre for Mathematical Biology at the University of Oxford's Mathematical Institute and lead author of the study.

One of the greatest challenges in cancer treatment is maximizing the impact of drug treatment on patients. Traditional treatment strategies focus on killing as many cells as possible and are based on “maximum tolerated dose” (MTD) therapy, where patients receive high doses of drugs continuously without interrupting treatment. However, these treatments are often ineffective against metastatic cancer due to the emergence of drug resistance.

Adaptive therapy strategies, which dynamically adjust treatment to reduce the rise in treatment-resistant patients, have emerged as a promising alternative, but their effectiveness is limited by the lack of a personalized approach that takes into account patient heterogeneity.

A new study published in 2011 found that Cancer ResearchResearchers from the University of Oxford and Moffitt Cancer Center in Florida present a novel framework that applies deep reinforcement learning (DRL) (a type of AI) to create adaptive treatment schedules for each patient with prostate cancer. Results show that this novel adaptive approach can double the time to recurrence compared to MTD or non-individualized treatment breaks.

First author Kit GallagherA PhD student at the University of Oxford's Mathematical Institute, he trained a DRL network on synthetic data from a mathematical model of prostate cancer. Reproduce behaviors seen in previous clinical trialsThis mathematical model was essential for generating sufficient amounts of “virtual patient” data, allowing researchers to evaluate treatment schedules that could not easily be tested clinically.

In our computational simulations, these schedules consistently outperform clinical standard-of-care protocols and common adaptive therapies for cancer treatment, demonstrating how these results can be translated to support clinical decision-making.

Kit Gallagher, Mathematical Institute, University of Oxford.

Results show that the DRL framework consistently outperformed traditional MTD and adaptive strategies used in clinical practice, delaying recurrence for all patients in the test cohort and more than doubling the time to recurrence in some patients. The DRL framework is also robust to changes and uncertainties in both patient treatment response and treatment intervals, which are crucial for applying this approach in the real world.

Importantly, the researchers demonstrated that they could extract interpretable treatment strategies from a “black box” deep learning network in a way that clinicians can understand and prescribe to patients.

“Interpretability has long been a major obstacle to incorporating machine learning approaches into clinical practice,” said Kit Gallagher. “If these frameworks are a black box and we can't understand how they derive treatment recommendations, we can't confidently apply them in the clinic. However, our new study shows that this obstacle is surmountable.”

This approach can also be applied to patients starting a new medication, who may not have a history of response to treatment and therefore cannot develop an individualized treatment schedule. In such cases, the researchers suggest a pathway where the patient would first begin a standard treatment cycle. A “virtual twin” of this patient would then be created based on data from this initial treatment, allowing the DRL model to be fine-tuned to create an individualized treatment schedule.

The researchers plan further studies to refine the method and explore its application to other types of cancer.

A study titled “Mathematical model-driven deep learning enables personalized adaptive treatments” has been published. Cancer ResearchThis research is supported by the National Cancer Institute.