Project: AI model for breast cancer detection and classification in histopathology

Today, there is a lot of variability in the assessment of images because we humans judge differently. AI should be able to find information in the images that we cannot detect.
We are developing a breast cancer detection and grading system (submitted manuscript) based on AI image analysis. We have created digital histopathology images by digitizing microscopy slides from the hospital archives in Stockholm. All images undergo careful quality assurance, processing and normalization. To date, we have generated digital images from more than 5000 breast cancer patients. Through careful analysis by a pathologist, a breast cancer can be divided into three groups: grade 1 with the lowest aggressiveness, grade 3 with the highest aggressiveness, and grade 2 as a gray zone in between. Patients with grade 3 tumors generally benefit from additional chemotherapy. We have created an AI model that can detect cancer in the images and stratify intermediate grade (NHG2) tumors to high or low grade. As NHG2 accounts for 50% of all breast cancers, this could have important clinical implications. This is a clinically important task that can help reduce gray areas and identify patients who need additional chemotherapy after surgery. We have also generated gene-specific models to predict spatial mRNA expression in breast cancer images. This has the potential to predict biomarker expression as well as molecular subtypes. In addition, similar methods have the capacity to replace more expensive molecular analyses in the future.