A groundbreaking study conducted at Northwestern Medicine suggests that an artificial intelligence (AI) tool could spare breast cancer patients unnecessary chemotherapy treatments by more accurately predicting patient outcomes. The research, published today in Nature Medicine, shows that the AI tool outperformed the analysis of disease by expert pathologists.
While current practice for determining treatment of breast cancer patients is partially based on evaluation of the cancer cells from a tumor biopsy, the tool developed by the Northwestern team also included assessing patterns of non-cancerous cells which the study shows are also important in disease progression.
By evaluating both cancerous and non-cancerous cells, the AI evaluations identified patients who are currently classified as high-risk or intermediate risk but will become long-term survivors. Having this information available will allow physicians to reduce the duration or intensity of chemotherapy treatments. The study is the first of its kind to use AI to comprehensively evaluate both cancerous and non-cancerous cells of invasive breast cancer.
“Our study demonstrates the importance of non-cancer components in determining a patient’s outcome,” said corresponding study author Lee Cooper, associate professor of pathology at Northwestern University Feinberg School of Medicine. “The importance of these elements was known from biological studies, but this knowledge has not been effectively translated to clinical use.”
Current practice for determining treatments for breast cancer patients involves grading by a pathologist who determines how abnormal the tissue appears. This process has remained largely unchanged for decades.
But previous research has shown the importance of other non-cancerous cells such as immune cells and cells that provide structure to tissue in determining the path of the disease.
Cooper’s team at Northwestern sought to build an AI model of breast cancer using digital images of cells that take into account the interactions of the cancerous and non-cancerous elements of breast cancer. AI and its ability to learn from and recognize complex patterns is a vital tool in this analysis since many of those patterns are difficult to recognize and evaluate using the human eye.
To build and train the AI model for the study, the investigators collaborated with the American Cancer Society which created a unique dataset of cancer patients derived from previous cancer prevention studies. The data had patient representation from 423 counties in the U.S. Many of the patients had received their diagnosis or care at community medical centers—an important distinction since many cancer studies are typically performed at academic medical centers that represent only a small slice of the overall diversity of the country’s population.
To train the model, the investigators gathered hundreds of thousands of human-generated annotations of cells and tissue structures within digital images through a global network of medical students and pathologists. These volunteers provided data over several years to enable the AI model to reliably interpret images of breast cancer tissue.
For the next phase of their work, the Northwestern researchers will use the AI tool prospectively on new patients in order to validate it for clinical use. The research also coincides with the transition, over the next three years, to a digital pathology platform at Northwestern Medicine for cancer diagnosis.
Further refinement of the model is also underway with an eye toward models appropriate for specific types of breast cancer such as triple-negative breast cancer and HER2-positive breast cancer, as cellular patterns for these may differ from one another.
