A new AI (artificial intelligence) tool can predict treatment outcomes for breast cancer patients with more precision, saving patients from unnecessary chemotherapy. That's according to a new study from Northwestern Medicine Research Institute. AI’s assessment of patient tissue better predicts a patient’s future course than an expert pathologist’s assessment.
Artificial intelligence tools can identify breast cancer patients who are currently classified as high or moderate risk but become long-term survivors. This means their chemotherapy can be shorter or less intense. This is important because chemotherapy can cause unpleasant and harmful side effects, such as nausea or, more rarely, damage to the heart.
Currently, pathologists determine treatments by evaluating cancer cells in a patient's tissue. But research shows that patterns of non-cancerous cells are important in predicting outcomes. This is the first study to use artificial intelligence to comprehensively assess cancer cells and non-cancerous cells in invasive breast cancer.
"Our study demonstrates the importance of non-cancerous components in determining patient prognosis," said Lee Cooper, associate professor of pathology at Northwestern University Feinberg School of Medicine and corresponding author of the study. "Biological research already knows the importance of these elements, but this knowledge has not yet been effectively translated into clinical applications."
The research will be published today (November 27) in the journal Nature Medicine.
In 2023, approximately 300,000 American women will be diagnosed with invasive breast cancer. About one in eight American women will be diagnosed with breast cancer in her lifetime.
During the diagnostic process, a pathologist will review the cancerous tissue to determine how abnormal the tissue is. This process, called staging, focuses on the appearance of cancer cells and has remained largely unchanged for decades. The grade determined by the pathologist helps determine what treatment the patient will receive. Many studies of breast cancer biology have shown that non-cancer cells, including immune system cells and cells that provide shape and structure to tissues, play important roles in maintaining or inhibiting cancer growth.
Cooper and colleagues built an artificial intelligence model to evaluate breast cancer tissue from digital images, measuring the appearance of cancer cells and non-cancerous cells and how they interact with each other.
"It's challenging for pathologists to evaluate these patterns because it's difficult for the human eye to reliably classify them," said Cooper, a member of Northwestern University's Robert H. Lurie Comprehensive Cancer Center. "The AI model measures these patterns and presents the information to the pathologist in a way that makes the AI's decision-making process clear to the pathologist." "
The artificial intelligence system analyzes 26 different attributes of the patient's breast tissue to generate an overall prognostic score. The system is also able to generate individual scores for cancer cells, immune cells and stromal cells so that the overall score can be interpreted to pathologists. For example, for some patients, a good prognostic score may be due to the properties of their immune cells, while for others, a good prognostic score may be due to the properties of their cancer cells. The patient's care team can use this information to develop a personalized treatment plan.
Employing this new model could provide patients diagnosed with breast cancer with more accurate risk estimates associated with their disease, empowering them to make informed decisions about their clinical treatment. In addition, the model can help assess treatment response, escalating or downgrading treatment based on changes in the microscopic appearance of tissue over time. For example, the tool might be able to identify how effectively a patient's immune system is targeting cancer during chemotherapy, allowing it to be shortened or less intensive.
"We also hope this model will reduce inequalities in patients diagnosed in community settings. These patients may not have access to a specialist breast cancer pathologist, and our AI model can help general practice pathologists evaluate breast cancer," Cooper said.
The study was conducted in collaboration with the American Cancer Society (ACS), which has created a unique dataset of breast cancer patients through its Cancer Prevention Research. The dataset represents patients from more than 423 counties in the United States, many of whom were diagnosed or treated at community health centers. This is important because most studies typically use data from large academic medical centers, which represent only a portion of the U.S. population. In this collaboration, Northwestern University developed the artificial intelligence software, while scientists from the American Cancer Society and the National Cancer Institute provided expertise in breast cancer epidemiology and clinical outcomes.
To train the AI model, scientists generate hundreds of thousands of human-generated annotations of cells and tissue structures in digital images of patient tissue. To do this, they created an international network of medical students and pathologists from several continents. These volunteers made this data available through the website over several years, allowing the artificial intelligence model to reliably interpret images of breast cancer tissue.
Next, the scientists will prospectively evaluate this model to validate its clinical use. This coincides with Northwestern Medicine's transition to using digital images for diagnosis over the next three years.
Scientists are also working to develop models for more specific types of breast cancer, such as triple-negative or HER2-positive. There are several different types of invasive breast cancer, and important tissue patterns may differ between types.
"This will improve our ability to predict outcomes and will provide further insights into the biology of breast cancer," Cooper said.