AI predicts lung cancer tumor growth after radiation: NHS study

Lung cancer is an especially deadly disease: It far outstrips other types in its share of overall cancer deaths, and as many as one-third of patients with the most common form, non-small cell lung cancer (NSCLC), experience recurrence after treatment, leading to a survival rate of just 15%.

There’s plenty of room for improvement, then, with those numbers suggesting that catching tumor regrowth early could greatly improve outcomes for NSCLC patients. A group of U.K. researchers have developed a tool that may do just that.

In a study newly published in the journal eBioMedicine, the researchers—hailing from the Royal Marsden NHS Foundation Trust, the London-based Institute of Cancer Research and Imperial College London—describe how they built an artificial intelligence model that can predict lung cancer patient’s individual risk of recurrence within two years of radiation therapy.

The scientists began by applying a handful of machine learning algorithms to a retrospective dataset comprising more than 650 patients with NSCLC, with a median follow-up time of just over two years after treatment.

Each of the algorithms took a different set of patient data into account, allowing the researchers to narrow down which individual factors appear to have the greatest impact on an individual’s chance of recurrence, then build a new model based on only the most important factors. Ultimately, those included the patient’s tumor size and stage and the type and intensity of radiotherapy they underwent, as well as their smoking status, body mass index and age.

The resulting AI model was able to predict a patient’s two-year risk of recurrence more accurately than current methods, including the widely used TNM staging system.

“This is an important step forward in being able to use AI to understand which patients are at highest risk of cancer recurrence and to detect this relapse sooner so that re-treatment can be more effective,” said Richard Lee, M.B.B.S., Ph.D., chief investigator of the study. “Relapse is also a key source of anxiety for patients. Reducing the number of scans needed in this setting can be helpful and also reduce radiation exposure, hospital visits and make more efficient use of valuable NHS resources.”

Because the model relies on easily accessible clinical data, the researchers said it could be easily applied by healthcare providers across the U.K. and beyond. Next up, they’re planning to bring imaging data into the equation.

“The next phase of this study will test machine learning models using imaging data alone and in combination with clinical data. We hope to find out how our model, which is based on patient characteristics and the treatment they received, is influenced by imaging scan data,” said study lead Sumeet Hindocha, M.B.B.S.

They’ll also look to apply the algorithm beyond lung cancer, with Lee noting, “Our model used features specific to this disease, but by refining the algorithm, this technology could have much wider application.”