Though the five-year survival rate of all breast cancer types is about 90% and rising, because of its vast variety and the continued lack of a cure, the disease is still responsible for a significant share of all female cancer deaths.
To speed up the race to a 100% survival rate, Owkin, maker of an artificial-intelligence-powered clinical research platform, has developed a deep learning model that’s able to classify the risk a patient with early-stage, localized breast cancer will experience metastatic relapse within the next five years.
In a retrospective study conducted in tandem with French cancer research institute Gustave Roussy, Owkin’s AI identified whether there was a high or low risk that a patient’s cancer would return and spread beyond the breast with an area under the curve—or AUC, an indicator of how a model’s accuracy compares to random chance—of more than 80%, the standard AUC baseline of success. The results of the preliminary study were presented at the European Society of Medical Oncology’s annual meeting.
The model was trained to analyze not only stained and digitized histological slides of a patient’s tumor but also their clinical histories, spanning variables including age at surgery, tumor stage and size, number of positive nodes and nodules, and surgery type.
In its analysis of a pool of more than 1,400 patients treated at Gustave Roussy between 2005 and 2013 for localized breast cancer, Owkin’s AI achieved an AUC of 81%. In comparison, using the model to analyze only slide images, without factoring in clinical history, resulted in an AUC of just 77%, the same rate achieved by the standard Cox model to measure survival, which is based only on baseline clinical data.
The only specialized equipment a hospital needs to run the model is a slide scanner, commonly found in most pathology laboratories, making the AI a widely accessible, low-cost tool to predict breast cancer relapse risk. The researchers said they’re hoping that the model will both reduce unnecessary chemotherapy treatments and encourage high-risk patients to seek new and potentially more effective alternatives to chemo.
Next up, they’ll continue to validate the model. That’ll include using it to analyze the data of larger groups of breast cancer patients as well as improving its accuracy by identifying new biomarkers in the patients determined to be at high risk of metastatic relapse.
Since its founding in 2016, Owkin has raised nearly $75 million and built partnerships with life sciences heavyweights like Amgen. With that momentum, the New York City-based company has launched its namesake platform, designed to help researchers build and train machine learning and deep learning AI models that can be used in precision medicine.
Besides the breast cancer survival model currently in development, Owkin is building a handful of others also focused on improving cancer care. In June, it debuted an AI-based tool built in partnership with AP-HP Greater Paris University Hospitals that can predict genomic subtypes of pancreatic cancer. And just this week, Owkin and Cleveland Clinic unveiled a model trained to predict the outcome and survival of hepatocellular carcinoma patients after liver transplant.