AI researchers predict immunotherapy response using CT scans

For the first time, researchers have demonstrated that artificial intelligence can predict a patient’s response to immunotherapies by analyzing CT scan images for a personalized “radiomic signature,” without the need for a biopsy.

The signature is calculated by the level of lymphocyte infiltration of a tumor, with algorithms providing a predictive score of efficacy—according to the study published in The Lancet Oncology, which assessed CD8 cells as an imaging biomarker for anti-PD-1 and PD-L1 monotherapies at several different tumor sites.

Researchers from Gustave Roussy, CentraleSupélec, Inserm, Paris-Sud University and TheraPanacea, a CentraleSupélec spin-off specializing in AI, radiation oncology and precision medicine, aimed to demonstrate how immunologically richer tumor micro-environments can correlate with the chances of effective treatment, using noninvasive CT imaging.

Their retrospective study looked to validate the radiomic signature method genomically, histologically and clinically in 500 patients with solid tumors at different sites.

Using machine learning, the algorithm was taught to extract relevant information from CT scans from studies that also gathered genomic data. Based solely on images, the algorithm learned to predict what the genome might have revealed about the infiltration of the tumor by the immune system.

RELATED: Predicting immuno-oncology success in lung cancer

The signature was also tested and validated in other cohorts, including The Cancer Genome Atlas, according to a statement from Gustave Roussy cancer center.

To test its use in real-world situations, the algorithm evaluated CT scans performed before the treatment of participants in five phase 1 immunotherapy trials. Researchers found that patients with higher radiomic scores showed treatment effects at 3 and 6 months and had better overall survival.

In the future, a clinical study will refine the signature, evaluating it retrospectively and prospectively, enrolling a larger number of patients stratified by cancer type. In addition, researchers intend to integrate data from tissue analyses and other tests.