Gradient Health Breast Density

OVERVIEW

The project revolves around the task of labeling and classifying breast density in mammographic images, a critical component in the early detection of breast cancer. The project aims to develop an AI-based model that can accurately assess mammogram breast density, a factor that has been clinically recognized as a significant risk indicator for breast cancer. We integrate OHIF with Google Sheets to enable real time collaboration and labeling in both Sheets and the OHIF Viewer directly.

Mammogram images are X-ray pictures of the breast, and the radiographic density of the breast tissue reflects the relative amounts of fat, connective and glandular tissue. Four primary categories are generally used to classify breast density: almost entirely fatty, scattered areas of fibroglandular density, heterogeneously dense, and extremely dense. The project's goal is to accurately and consistently categorize these densities, overcoming the limitations of subjectivity and variability inherent in human interpretation.

The project begins with the acquisition of a large dataset of anonymized mammogram images. These images are then preprocessed to remove noise and improve clarity, creating a better foundation for the labeling process. The labeling is carried out by professional radiologists who categorize the images according to the four breast density classes. This annotated dataset serves as the ground truth for training our AI model.

In the subsequent phase, the labeled dataset is used to train a deep learning model to predict the density category of new mammogram images. The model could be based on convolutional neural networks (CNNs), which have shown excellent performance in image classification tasks. It will be trained and validated using the mammogram data, adjusting its parameters to optimize predictive accuracy.

The final part of the project is the model evaluation stage. Here, the AI model's performance is assessed using a separate test dataset to ensure the model's ability to generalize beyond the training data. Performance metrics, such as accuracy, sensitivity, and specificity, will be computed to evaluate the model's diagnostic power.

This medical imaging labeling project holds significant potential for improving breast cancer detection and diagnosis. By providing an accurate, consistent, and objective measure of breast density, it can aid radiologists in making informed decisions about follow-up procedures and treatment strategies. Furthermore, by establishing the foundations for automatic breast density classification, it opens up new avenues for large-scale breast cancer screening programs.