DeepTumour: Enhancing an Advanced Algorithm for Tumour Origin Identification

Improving DeepTumour, a computer program that will help doctors diagnose and treat cancer patients by identifying where a cancer started in the body

Project Summary

Objective: I am working on improving DeepTumour, a computer program that will help doctors diagnose and treat cancer patients by identifying where a cancer started in the body.

Impact and relevance to cancer: Accurately identifying where a cancer began and what type of cells it came from is crucial. This knowledge helps doctors pick the most effective treatment, ultimately improving patients’ chances for long-term survival. However, identifying the cancer type can be challenging as tumours with different origins can sometimes look very similar, while other cancers change appearance over time. This is particularly a concern for cancers of unknown primary (CUPs), which are cancers that have already spread by the time they cause symptoms and doctors cannot identify their starting point. It is also a problem in the case of multiple primary cancers (MPC), a situation in which the patient has had two or more cancers in the past and one of them has come back, but doctors don’t know which. Studies have shown that around 17% of cancer patients with MPC and 5% with CUPs could get a lot of help if we could figure out where their cancer started. In Canada, that is about 10,000 patients every year.

Previous research: Our team at the Ontario Institute for Cancer Research developed DeepTumour, an artificial intelligence system that can distinguish among 24 common cancer types with an accuracy of 88%. DeepTumour uses patterns of changes in cancer’s genes collected from whole-genome sequencing, a technique that provides a detailed view of cancer mutations and is becoming more commonly used in regular cancer care.

Project methods: My project aims to improve DeepTumour's performance by training it with data from Marathon of Hope Cancer Centres Network that contains cancer types new to DeepTumour. This will enable DeepTumour to classify more tumor types. I will also explore various computational methods to identify the most effective approach for prediction. Ultimately, I will integrate DeepTumour into an automated diagnostic system, aiding in the diagnosis and treatment of cancer patients, especially the ones with CUPs and MPC.

Quotes

“I am thrilled to receive this award and deeply grateful for this amazing opportunity! This achievement wouldn't have been possible without the unwavering support of my mentor, colleagues, and loved ones. With this award, I can further my passion for advancing DeepTumour, an artificial intelligence system designed to improve cancer identification and diagnosis, especially for challenging cases such as patients who develop two or more tumours simultaneously, and those with metastatic cancers whose origin is unknown.”

• Xindi Zhang, HI&DS Awardee

“Xindi's project will enhance DeepTumour, an artificial intelligence system for identifying what organ a metastatic tumour originated in. By applying machine learning to cancer genomic data, DeepTumour accurately identifies various common cancer types from mutation profiles. If successful, this tool will assist in diagnosing and personalizing treatments, especially for challenging cases like multiple simultaneous tumours and metastatic cancers of unknown origin, thereby improving patient outcomes and increasing survival rates for thousands of patients each year.”

• Dr. Lincoln Stein, mentor