Unsupervised Histopathology Search Engine for Diagnosis of Primary and Metastatic Cancers (CANCERCH)

In the realm of cancer diagnosis, a groundbreaking initiative is underway to revolutionize the way we identify primary and metastatic cancers. Harnessing the capabilities of Artificial Intelligence (AI), this project aims to create an Unsupervised Histopathology Search Engine! Imagine a powerful tool that can sift through massive digital archives of cancer images, guiding clinicians in distinguishing the origin of primary and metastatic cancers more efficiently.

Why is this Important?  Traditional methods rely on pathologists examining Hematoxylin & Eosin stained Whole Slide Images (WSIs) or ancillary testing, a time-consuming process prone to challenges. This project builds on recent advancements in AI to encode and analyze these images in an unsupervised manner. This not only enhances the speed of diagnosis but opens avenues for uncovering hidden patterns within cancer data.

How Does it Work? The proposed Histopathology Search Engine breaks down microscopic images into smaller patches and extracts cell and image characteristics from them. By deploying sophisticated AI models, it identifies similarities within the data without the need for explicit training and human supervision. The unsupervised exploration allows for the discovery of patterns across all cancers and within specific cancer types in a new way that have never been explored.

Potential Impact:

• Accurate Diagnosis of Rare Cancers: The tool's ability to identify and study rare metastatic cancers provides a valuable resource for understanding and treating less common forms of the disease.
• Treatment Tailoring: By linking histological features with clinical outcomes, the project aims to uncover novel subtypes that may respond differently to specific therapies, paving the way for more personalized treatment strategies.

Relevance to Cancer Research: This study focuses on the Personalized OncoGenomics (POG) trial dataset, providing a real-world application in collaboration with the BC Cancer Institute. The dataset spans 26 different cancers, offering a rich source of information to enhance our understanding of the relationships between histopathological features and clinical outcomes.

Quotes

"Metastatic cancers account for up to 90% of cancer deaths. However, determining the origin of these cancers (i.e., where the cancer originally started) is challenging. I am excited to use the Personalized OncoGenomic dataset from the BC Cancer Research Institute, supported by MOHCCN, and leverage AI to identify the site of origin for metastatic cancers. Through this collaboration, I hope we will contribute to advancing cancer research and improving patient care."

• Ali Khajegili Mirabadi, HI&DS Awardee
  

"Supported by a MOHCCN HI&DS Award, Ali's project aims to develop more objective methods for rare or metastatic cancer diagnosis, serving as a diagnostic adjunct to pathologists. We are excited to shift the current research paradigm by employing AI-driven computational pathology to identify potential biomarkers from tissue images. This could aid in accurately determining the site of origin for metastatic cancers and identifying potential new patient subgroups with distinct outcomes and responses to therapies."

• Dr. Ali Bashashati, mentor