Glossary
Overall survival: the time a patient lives, regardless of disease status.
News
Article
An artificial intelligence-based image analysis model has more accurately predicted cancer biomarkers and outcomes compared with conventional methods.
An AI-based image analysis model has more accurately predicted biomarkers and outcomes vs with conventional methods: © stock.adobe.com.
Among some patients with cancer, an artificial intelligence (AI)-based image analysis model has been shown to potentially be able to more accurately predict cancer biomarkers and patient survival when compared with conventional companion diagnostic methods, it has been announced.
The development was revealed in a study published in the journal Communications Medicine and detailed in a news release issued by the next generation AI techbio and precision medicine company Caris Life Sciences, the company behind the image analysis model.
“This research represents a notable step forward in employing AI to analyze oncological biomarkers, highlighting its substantial potential to improve patient treatment outcomes,” researchers wrote in Communications Medicine.
“In conclusion, our study introduces a framework that transcends conventional pathology slide analysis,” researchers ultimately concluded in the study. “We have demonstrated the capability of our model to predict critical prognostic biomarkers for immunotherapy, highlighting its potential as a supportive tool for pathologists. The heatmaps generated by our model not only facilitate the identification of impactful patches or regions of interest but also foster a collaborative environment where pathologists can interpret AI findings. This synergy could uncover new diagnostic patterns and paradigms, potentially revolutionizing patient stratification and prognosis. Ultimately, we envision AI tools delivering multifaceted support to contemporary and future clinical practices.”
Overall survival: the time a patient lives, regardless of disease status.
Caris’ AI model, as part of the study, analyzed the data of more than 35,000 patients. As detailed in the news release, among patients with breast cancer the model scored PD-L1-positive phenotype status, meaning the level and location of PD-L1 protein expression within a tumor, assessing overall survival of patients treated with Keytruda (pembrolizumab). Under the AI model, patients identified as likely to respond to Keytruda had approximately half the risk of death compared to those not identified as responders, whereas traditional PD-L1 scoring were found to have a much smaller impact on predicting which patients would survive longer.
Furthermore, among patients with colorectal cancer, AI methods predicted mismatch repair deficiency and microsatellite instability — genetic characteristics that influence a tumor's behavior and response to treatment — equal to traditional scoring methods.
“Traditional PD-L1 testing can undercall positive cases, especially near the 1% threshold,” said Dr. Matthew Oberley, Chief Clinical Officer and Pathologist-in-Chief at Caris, in the news release. “Caris’ AI model enhances predictive accuracy, integrating features from both staining methods, and exhibits superior prognostic precision compared to current biomarker assessments. Clinical adoption of this tool could improve the precision and efficiency of cancer patient evaluation and aid clinical decision making.”
“This study highlights how AI can significantly improve the accuracy and efficiency of tissue sample evaluation, and down the line, this has the potential to guide immunotherapy decisions and enhance patient outcomes,” said Dr. George W. Sledge, Jr., Caris EVP and Chief Medical Officer, in the news release.
Artificial intelligence has the potential to transform the cancer treatment and management landscape, as experts who previously spoke with CURE have explained.
“Right now, AI is definitely being folded into different levels of the entire journey for patients,” says Dr. Soroush Rais-Bahrami of Wake Forest University School of Medicine in Winston-Salem, North Carolina.
“[AI] has a complementary role, as far as it can make our jobs easier and [make] sure that things are not falling through the cracks,” says Richard Boyajian.
An advanced practice registered nurse and nurse practitioner in the Department of Radiation Oncology with the Dana-Farber Brigham Cancer Center in Boston, Boyajian founded the Virtual Prostate Cancer Clinic at Brigham and Women’s Hospital in 2016. He is also a member of the CURE advisory board.
“The quicker you can present all the data succinctly to the provider, they’re able to make a decision based on these multiple points that should be considered,” Boyajian says. “… There may be other things that are in play that AI can calculate to build the picture so we can put whatever’s going on into a context, if you will.”
For more news on cancer updates, research and education, don’t forget to subscribe to CURE®’s newsletters here.