Managing Editor, OncLive®
Kristi Rosa joined MJH Life Sciences in 2016 and has since held several positions within the company. She helped launch the rapidly growing infectious disease news resource Contagion, strengthened the Rare Disease Report, of HCPLive, and now serves as the main digital news writer for OncLive. Prior to working at the company, she served as lead copywriter and marketing coordinator at The Strand Theater. Email: firstname.lastname@example.org
November 19, 2020 - A computational method comprised of clinicodemographic variables with deep learning of pretreatment histology images may be able to effectively predict response to immunotherapy in patients with advanced melanoma.
A computational method comprised of clinicodemographic variables with deep learning of pretreatment histology images may be able to effectively predict response to immunotherapy in patients with advanced melanoma, according to findings from a study published in Clinical Cancer Research.1
“While immune checkpoint inhibitors have profoundly changed the treatment landscape in melanoma, many tumors do not respond to treatment, and many patients experience treatment-related toxicity,” corresponding study author Iman Osman, MD, medical oncologist in the Departments of Dermatology and Medicine (Oncology) at New York University (NYU) Grossman School of Medicine and director of the Interdisciplinary Melanoma Program at NYU Langone’s Perlmutter Cancer Center, stated in a press release.2 “An unmet need is the ability to accurately predict which tumors will respond to which therapy. This would enable personalized treatment strategies that maximize the potential for clinical benefit and minimize exposure to unnecessary toxicity.”
In the study, investigators utilized computer algorithms referred to as deep convolutional neural networks (DCCN) to examine digital images of metastatic melanoma tumors and determine patterns linked with response to treatment with immune checkpoint blockade. Using this approach, they created a response classifier which was designed to predict if a patient’s tumor would respond to this treatment approach or would progress after treatment.
Results demonstrated that the multivariable classifier predicted response with AUC of 0.800 on images from the slide scanner Aperio AT2 and area under the curve (AUC) of 0.805 on images from the Leica SCN400. Moreover, the classifier was found to effectively stratify patients into groups who were at high or low risk for progressive disease.
Moreover, patients who were determined to be at high risk for progression were found to have significantly worse progression-free survival (PFS) versus those who were considered to be low risk (P =.02 for the Aperio AT2 and P =.03 for the Leica SCN400).
“Several recent attempts to predict immunotherapy responses do so with robust accuracy but use technologies, such as RNA sequencing, that are not readily generalizable to the clinical setting,” corresponding study author Aristotelis Tsirigos, PhD, professor in the Institute for Computational Medicine at NYU Grossman School of Medicine and member of NYU Langone’s Perlmutter Cancer Center, stated in the release. “Our approach shows that responses can be predicted using standard-of-care clinical information such as pretreatment histology images and other clinical variables.”
For the study, investigators examined data from a training cohort of 121 patients with metastatic melanoma who had received treatment with immunotherapy between 2004 and 2018. All of these patients had received frontline treatment with a CTLA-4 inhibitor, a PD-1 inhibitor, or a combination of the 2 agents. The clinical outcomes for these patients included disease progression or response to treatment; those who had achieved stable disease were excluded from the analysis.
The investigators validated the DCCN response classifier in an independent cohort of 30 patients with metastatic melanoma who had received treatment at Vanderbilt-Ingram Cancer Center between 2010 and 2017. They examined the performance of this classifier by calculating the AOC, where a score of 1 correlates with perfect prediction with the model. The model was found to have an AUC of about 0.7 in the training and validation cohorts.
Multivariable logistic regressions were then conducted, which combined the DCCN prediction with conventional clinical features. To augment prediction accuracy, data regarding the DCCN prediction, ECOG performance status, and treatment regimen were all incorporated into the final model. In both cohorts examined, the multivariable classifier achieved an AUC rose by 0.1. Now, the model was able to stratify patients based on risk for progression.
Additional data revealed that most of the patients, or about 64%, in the training group had received treatment with a single-agent CTLA-4 inhibitor. A little more than half of the patients, or about 53%, in the validation group had received PD-1 inhibitors. This information indicates that some of the predictive patterns observed are not necessarily dependent on the immune checkpoint target. Moreover, data gleaned from class activation mapping showed that cell nuclei played a central role in the DCCN predictions in that more and larger nuclei was determined to be associated with disease progression.
The study was not without limitations. For one, a relatively small number of images were utilized to train the computer algorithm; only 302 images and 40 images were examined in the training and validation cohorts, respectively. The authors admit that thousands of images may be required to adequately train models for clinical-grade performance.
“There is potential for using computer algorithms to analyze histology images and predict treatment response, but more work needs to be done using larger training and testing datasets, along with additional validation parameters, in order to determine whether an algorithm can be developed that achieves clinical-grade performance and is broadly generalizable,” concluded Tsirigos.