'Radiomics' Predicts Chemo Response in NSCLC

Roxanne Nelson, RN, BSN

April 01, 2019

A new approach combining CT and "radiomics," which extracts data from medical images, may be able to determine which patients with lung cancer are most likely to respond to chemotherapy.

Platinum-based chemotherapy is the current standard of care for first-line treatment of advanced-stage non–small cell lung cancer (NSCLC) in patients with no actionable mutations, but only about a quarter of patients will respond. There is currently no effective way of predicting which patients will benefit most from chemotherapy, as no clinically validated biomarkers have been identified.

Radiomics may help. The approach involves quantifying tumor characteristics based on imaging and advanced bioinformatics tools.

In a new study, researchers found that using CT imaging of radiomic features from within and outside the lung nodule could predict time to progression and overall survival, as well as response to chemotherapy in patients with NSCLC.

The study was published in the journal Radiology: Artificial Intelligence.

Study author Anant Madabhushi, PhD, professor of biomedical engineering and director of the Center for Computational Imaging and Personalized Diagnostics at Case Western Reserve University in Cleveland, Ohio, commented that following validation of the image classifier, they anticipate that the test could become a companion diagnostic for patients with lung cancer scheduled to receive chemotherapy with pemetrexed (Alimta, Lilly).

"The oncologist would likely order the test and it would run off a standard diagnostic CT scan," he told Medscape Medical News. "The actual deployment of the test could be through a cloud-based platform where the scan could be pushed up into a HIPAA-compliant cloud framework and the analysis done there and the results of the analysis then securely emailed back to the ordering oncologist."

A tech working in radiology or oncology could then use a simple web interface to push the scan into the cloud, Madabhushi explained. "Since all the analysis would be done remotely, no technical expertise would be needed on the ground or in the clinic."

Based on the results of the test, the ordering oncologist might then consider either switching to another agent or using combination therapy (eg, chemoradiation) or possibly staying with the original treatment regimen.

Study Details

To identify the role of radiomics texture features both within and outside the nodule in predicting response to chemotherapy and outcomes in NSCLC, Madabhushi and colleagues conducted a retrospective study using data from 125 patients who had been treated with pemetrexed-based platinum doublet chemotherapy at Cleveland Clinic.

The cohort was randomly assigned into two sets with an equal number of responders and nonresponders in the training set. The training set was comprised of 53 patients and the validation set had 72 patients.

The authors trained a machine-learning classifier with radiomic texture features extracted from intra- and peritumoral regions of non–contrast-enhanced CT images, and that was used to predict response to chemotherapy. The radiomic risk-score signature was then generated by using the "least absolute shrinkage and selection operator with the Cox regression model."

Their analysis showed that several of the radiomic features, combined with a quadratic discriminant analysis classifier, showed a mean maximum area under the receiver operating characteristic curve (AUC) of 0.82 ± 0.09 (standard deviation) in the training set, and a corresponding AUC of 0.77 in the independent testing set.

The radiomics signature was also significantly associated with time to progression (hazard ratio [HR], 2.8; 95% confidence interval [CI]: 1.95, 4.00; P < .0001) and overall survival (HR, 2.35; 95% CI: 1.41, 3.94; P = .0011).

Overall, the radiomics model had the highest net benefit as far as predicting which high-risk patients should receive treatment, as compared with a clinicopathologic model and simple strategies such as follow-up of all patients or no patients.

Broader Applications

Madabhushi noted that this approach has potential use in all patients with and not just those who are candidates for chemotherapy. 

"Our current published work is on lung cancer patients undergoing chemotherapy, but we have already generated data in other studies showing that the approach can be used to predict response in the context of lung cancer patients receiving immunotherapy," he said. "We hope to be able to publish those results soon."

He added that they are continuing to validate this model. "As recent studies have shown, it is critical to independently validate AI models on data from multiple different sites and institutions," he said. "We feel confident about our classifier, but seek to demonstrate that the approach is robust to inter-site differences. Performing this rigorous multi-site validation is an important prerequisite to clinical deployment of AI models, especially when it comes to predicting therapeutic response."

The study was supported by the US Department of Defense, the National Cancer Institute, the National Institute of Diabetes and Digestive and Kidney Diseases, the National Center for Research Resources, the Ohio Third Frontier Technology Validation Fund, the Wallace H. Coulter Foundation Program in the Department of Biomedical Engineering at Case Western Reserve University (CWRU), and the Clinical and Translational Science Award Program at CWRU. Madabhushi has disclosed no relevant financial relationships.

Radiology: Artificial Intelligence. Published online March 20, 2019. Full text

For more from Medscape Oncology, join us on Twitter and Facebook


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: