AI Case Study

Scientists at Institute Gustave Roussy predict immunotherapy efficacy in patients with machine learning

Scientists at Cancer Research Institute Gustave Roussy in France and four other institutions have designed an algorithm to analyse CT scan images of tumors. By observing the level of white blood cells working towards fighting the tumor, the system can predict the effectiveness of immunotherapy.



Healthcare Providers And Services

Project Overview

"A new study finds that artificial intelligence can process medical images to extract biological and clinical information to aid immunotherapy treatment.

By designing an algorithm and developing it to analyze CT scan images, researchers at cancer research institute Gustave Roussy in France and four other institutions have created a “radiomic signature,” which is the extraction of large numbers of features from medical images.

This signature defines the level of lymphocyte infiltration of a tumor where white blood cells leave the bloodstream and migrate toward the tumor to fight it. This then generates a predictive score for the effectiveness of immunotherapy.

Presently, no marker can accurately identify patients who will respond to anti-PD-1/PD-L1 (proteins) immunotherapy in a situation where only 15 percent to 30 percent of patients will respond. The greater the presence of lymphocytes, the greater the chance that immunotherapy will be effective, according to scientists in the study published in The Lancet Oncology.

With the use of machine learning technology, researchers taught the algorithm that they had designed to analyze CT scan images to use relevant information extracted from scans that held tumor genome data.

“Thus, based soley on images, the algorithm learned to predict what the genome might have revealed about the tumor immune infiltrate, in particular with respect to the presence of cytotoxic T-lymphocytes in the tumor, and it established a radiomic signature,” according to the study.

Additional tests of the signature found that patients in whom immunotherapy was effective at 3 and 6 months had higher scores and better overall survival rates.

Now, a forthcoming additional study will assess the radiomic signature retrospectively and prospectively, and will involve more patients stratified according to cancer type to refine the signature. “This will also employ more sophisticated automatic learning and artificial intelligence algorithms to predict patient response to immunotherapy,” according to the study. “To that end, the researchers are intending to integrate data from imaging, molecular biology and tissue analysis.”

Reported Results

"The results, from a French research institute, are important because the research suggests that AI can be used to create a predictive score on the efficacy of immunotherapy in a patient, thus saving time and increasing the chances for success in treatment."


"A linear elastic-net model was used as the regression method by use of the R package glmnet (version2.0–10)forfeatureselectionandmodelbuilding.30 The regularisation parameter λ was de ned by use of cross-validation and the α penalty was set to 0·5 after a grid search.Themachine-learningalgorithm—ie,theradiomic score—provides a mathematical formula that predicts the abundance of CD8 cells by use of the gene expression signature with imaging data with the equation:
ˆy=a1X1 +a2X2 +...+aiXi +b

where ŷ is the radiomic score, ai is the coe cient of the variable i, Xi is the value of the variable i determined fromtheinputimage,andbistheintercept.Asensitivity analysis of the e ects of the non-radiomic features on the performance of the signature was also done." (paper)


R And D

Core Research And Development


"Immunotherapy is the process of using the body’s own immune system to fight the cancer. The goal is for physicians to be able to use imaging to identify biological phenomena in a tumor located in any part of the body without having to perform a biopsy." (healthdatamanagement)

"Because responses of patients with cancer to immunotherapy can vary in success, innovative predictors of response to treatment are urgently needed to improve treatment outcomes. We aimed to develop and independently validate a radiomics-based biomarker of tumour-infiltrating CD8 cells in patients included in phase 1 trials of anti-programmed cell death protein (PD)-1 or anti-programmed cell death ligand 1 (PD-L1) monotherapy. We also aimed to evaluate the association between the biomarker, and tumour immune phenotype and clinical outcomes of these patients." (thelancet)



"Input variables for the machine-learning method consisted of 84 variables: 78 radiomic features, ve locations (labelled as binary variables), and one global imaging variable, the peak kilovoltage, given its established e ect on radiomic output." (paper)