Statistical Inference in Computational Pathology

Breast cancer is a prevalent disease affecting millions of women worldwide, and early detection is crucial for effective treatment and improved survival rates. Histopathology image analysis is a common tool in the diagnosis and management of breast cancer. This dissertation discusses the current state of breast cancer image analysis, with a focus on the development of statistical frameworks and computational pathology. Two common types of histopathology images are whole slide images and tissue microarray cores. These images are typically large in size and have a widely varying number of images and pixels across patients and trials. Furthermore, inconsistent stain colors across different datasets can produce cross-dataset variability. To handle these challenges, the importance of Multiple Instance Learning (MIL) is emphasized, as well as the potential benefits of domain adaptation approaches. The essential role of data preparation and preprocessing in these systems is also highlighted, along with the potential of these tools for treatment effect estimation and Natural Language Processing (NLP) for medical text data. The potential benefits of image analysis for breast cancer diagnosis such as increased classification performance and improved domain transferability are reviewed. We also discuss the challenges and limitations including the need for large datasets and high computational power. Despite these challenges, the continued development and advancement of breast cancer image analysis holds great promise for improving outcomes for those affected by the disease.