Artificial intelligence can predict the response to immune checkpoint blockade in patients with metastatic melanoma

A computational method that combines clinicodemographic variables with deep learning of pre-treatment histological images could predict the response to immune checkpoint blockade among patients with advanced melanoma, according to results published in Clinical Cancer Research, a journal of the American Association for Cancer Research.

“While immune checkpoint inhibitors have profoundly changed the melanoma treatment landscape, many cancers do not respond to treatment and many patients experience treatment-related toxicity,” said 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. “An unmet need is the ability to accurately predict which cancers will respond to which therapy. This would allow for personalized treatment strategies that maximize clinical benefit potential and minimize exposure to unnecessary toxicities.”

“Several recent attempts to predict immunotherapy responses do so with robust accuracy, but use technologies, such as RNA sequencing, that are not easily generalizable to the clinical setting,” said study author Aristotelis Tsirigos, Ph. D., a professor at the Institute of Computational Medicine at NYU Grossman School of Medicine and a member of NYU Langone’s Perlmutter Cancer Center. “Our approach shows that responses can be predicted using standard-of-care clinical information such as pre-treatment histological images and other clinical variables.”

The researchers used data from a training cohort of 121 metastatic melanoma patients who received immune checkpoint blockade treatment between 2004 and 2018. All patients were treated with first-line anti-CTLA-4 therapy. , anti-PD-1 therapy or a combination of both, and clinical outcomes were recorded as disease progression or response, which included complete or partial responses (patients with stable disease were excluded for this proof of principle study) . The researchers used computer algorithms called deep convolutional neural networks (DCCNs) to analyze digital images of metastatic melanoma cancers and identify patterns associated with treatment response. Through this approach, they developed a response classifier, which aimed to predict whether a patient’s untreated tumor would respond to immune checkpoint blockade or progress after treatment. This DCCN response classifier was validated in an independent cohort of 30 metastatic melanoma patients treated at the Vanderbilt-Ingram Cancer Center between 2010 and 2017.

The performance of the DCCN response classifier was evaluated by calculating the area under the curve (AUC), a measure of the accuracy of the model, where a score of 1 corresponds to the perfect prediction. The DCCN prediction model achieved an AUC of approximately 0.7 in both the training and validation cohort.

To increase the accuracy of the model prediction, the researchers performed multivariable logistic regressions that combined DCCN prediction with conventional clinical features. The final model included DCCN prediction, Eastern Cooperative Oncology Group (ECOG) performance status, and treatment regimen (anti-CTLA-4 monotherapy, anti-PD-1 monotherapy, or combination therapy). In both training and validation cohorts, the multivariable classifier achieved an AUC around 0.8. In the validation cohort, the classifier could stratify patients high versus low risk of disease progression, with significantly different progression-free survival outcomes between the two groups.

While the majority of patients in the training cohort received anti-CTLA-4 monotherapy (approximately 64% of patients), the majority of patients in the validation cohort received anti-PD-1 agents (approximately 53 % of patients). The results suggested that some predictive models are not specific to the immune checkpoint target, Osman noted. Class activation mapping, which can identify regions within digital images that the neural network uses to generate predictions, suggested that cell nuclei were important for DCCN predictions, where larger and more numerous nuclei were correlated. with the progression of the disease. “These results suggest that ploidy could be one of the biological determinants detected by DCCN,” he added.

“There is the potential to use computer algorithms to analyze histological images and predict response to treatment, but more work needs to be done using larger training and testing datasets, along with additional validation parameters, in order to determine if it is It is possible to develop an algorithm that achieves clinical outcomes improve performance and is broadly generalizable, “Tsirigos said.

Limitations of the study include the relatively small number of images used to train the computer algorithm, which included 302 images in the training cohort and 40 images in the validation cohort. “There is data to suggest that thousands of images may be needed to train models that achieve clinical-level performance,” Tsirigos said.