Dr Phedias Diamandis
Assistant Professor, University of Toronto
Brain tumours represent a diverse group of disease with different treatments and outcomes for each tumour type. For the past decade, tumour diagnosis has been conducted based on looking and analysing a tumour sample under a microscope. While molecular testing is improving the accuracy of tumour diagnosis, visual analysis continues to play an important role in determining diagnosis. However, visual examination is a subjective process and has a significant amount of differences depending on the individual viewing the sample.
Preliminary research has shown that there is a growing interest in using artificial intelligence (AI) to improve brain tumour diagnosis. However, studies so far have largely focused on relatively niche tasks using pre-defined samples, which limits its use.
To address this, the research team led by Dr Diamandis have developed a brain tumour classification tool by using an emerging form of AI known as convolutional neural networks (CNNs). The aim of this research project is to “train" the classification tool to differentiate the different types of brain tumours. The training involves exposing the classification tool to a large series of images with known diagnoses, survival, and therapy response. This exposure will allow the classification tool to “learn" even the smallest of patterns associated with certain clinical events. The classification tool will then allow researchers to predict tumour behaviour and response to treatment.
This classification system will have a significant clinical impact, as it will improve diagnostic accuracy and speed, as well as allow patients to receive treatment regimens tailored to their specific tumour type.