AI Model Successfully Predicts Patients' Sensitivity to Cancer Therapies, Study Finds

Artificial intelligence (AI) could help predict responses to non-small cell lung cancer systemic therapies, according to a study published in Clinical Cancer Research.

Researchers used standard-of-care computed tomography (CT) images of patients diagnosed with non—small cell lung cancer (NSCLC) to define radiomics signatures. Radiomics depends on quantitative transformations of images into datasets to enable “high-throughput data mining and automated analysis of patterns present in images,” researchers said. The signatures in this study were able to predict the sensitivity of tumors to nivolumab, docetaxel, and gefitinib.

Selecting patients for targeted therapies or immunotherapy is important so that individuals are matched to the treatment most likely to benefit them. However, for patients with NSCLC, therapy personalization relies on pretreatment biomarkers that are acquired in a tumor biopsy. Performing the necessary genomic analyses “are typically limited to a single biopsy sample, are difficult to perform repeatedly, and thus cannot capture the spatial and temporal heterogeneity of disease,” researchers said.

Researchers prospectively collected data and retrospectively analyzed the data among participants in multicenter clinical trials. Of the patients included, 92 received nivolumab, 50 received docetaxel, and 26 received gefitinib. Patients were randomly divided into training or validation cohorts and 1160 radiomics features were extracted from the largest measurable lung lesion of each patient.

“Radiologists' interpretation of CT scans of cancer patients treated with systemic therapies is inherently subjective,” said Laurent Dercle, MD, PhD, a lead author of the study. He continued, “The purpose of this study was to train cutting-edge AI technologies to predict patients' responses to treatment, allowing radiologists to deliver more accurate and reproducible predictions of treatment efficacy at an early stage of the disease."

Models created to predict treatment sensitivity in the training cohort predicted a score between 0 (highest treatment sensitivity) and 1 (highest treatment insensitivity). The scores were based on the largest measurable lung lesion identified at baseline for each patient. Eight radiologic features, including changes in tumor volume, heterogeneity, shape, and margin, were used to build 3 prediction models.

When calculating the area under the curve (AUC) to measure each model’s accuracy (where 1 is the most accurate), nivolumab had an AUC of .77, docetaxel had an AUC of .67, and gefitinib of .82.

Because the study used small sample sizes, Dercle notes future studies should include AI use on larger patient datasets. “"With AI, cancer imaging can move from an inherently subjective tool to a quantitative and objective asset for precision medicine approaches,” Dercle said.

Reference

Dercle L, Fronheiser M, Lu L, et al. Identification of non-small cell lung cancer sensitive to systemic cancer therapies using radiomics [published online March 20,2020]. Clin Cancer Res. doi: 10.1158/1078-0432.CCR-19-2942.