Image-Based Model Predicts Radiation Tx Failure in Lung Cancer

An image-based deep learning framework can predict radiation treatment failure in lung cancer patients, according to a study published in the July issue of The Lancet Digital Health.

Bin Lou, Ph.D., from Siemens Healthineers in Princeton, New Jersey, and colleagues identified radiation sensitivity parameters that can predict treatment failure in patients with primary or recurrent lung cancer, and patients with solitary metastases or oligometastases to the lung treated with stereotactic body radiotherapy. A total of 849 patients were included in the internal study cohort, and 95 were included in an independent validation cohort. Pre-therapy lung computed tomography (CT) images were input into Deep Profiler to generate an image fingerprint that predicts treatment outcomes.

The researchers found that radiation treatments failed at a significantly higher rate among patients with high Deep Profiler scores versus those with low scores (three-year cumulative incidence of local failure, 20.3 versus 5.7 percent in the internal study cohort; hazard ratio [HR], 3.64). Local failure was independently predicted by Deep Profiler (HR, 1.65). Treatment failure was predicted by models that included Deep Profiler and clinical variables with a concordance index of 0.72, which marked a significant improvement from classical radiomics or clinical variables alone. In the validation cohort, Deep Profiler performed well, predicting treatment failures across diverse clinical settings and types of CT scanners.

"The most important message in our study is that predictive features can be learned from CT images and can contribute to the individualization of radiation dose," the authors write.

Several authors disclosed financial ties to Bayer and Siemens; several authors are named inventors on a patent pending for use of Deep Profiler and iGray to personalize radiotherapy dose.

-----------------------------------------------------------------------
Editor's Note:

If you have any suggestion to the content, 

please email: Julia.Zhang@ubmsinoexpo.com