A paper that StFX computer science professor Dr. Jacob Levman co-authored on sample size considerations in machine learning applied to medical imaging has gained national acclaim, receiving The Editor’s Award of the Canadian Association of Radiologists Journal.
The paper is co-authored with Dr. Pascal Tyrrell of the University of Toronto.
Dr. Levman says he was pleased and surprised by the national award. “I had already noticed that despite the paper being quite recent it was already been cited on a regular basis, but I wasn't expecting it to receive top honours from a strong journal.”
Their paper, ‘Sample-Size Determination Methodologies for Machine Learning in Medical Imaging Research: A Systematic Review,’ looks at the scientific literature to assess sample size determination methods (i.e., how many samples do you need for your given application), for machine learning applied to medical imaging. It demonstrates that this area of research has a paucity of studies devoted thereto, Dr. Levman says.
“Machine learning technologies are heavily dependent on sample size, that is, how many examples are available from which the machine learns the underlying patterns it will rely upon to make predictions. So, for example, a machine to detect cancer from an image can be trained on 100 examples or 100,000 examples, or any other number,” Dr. Levman says.
“We know that the number of samples available for training greatly affects the performance and reliability of the technology that will be deployed. As such, this topic is extremely important towards the creation of reliable machine learning technologies.”
Dr. Levman says he hopes this work will spur additional interest and effort among the machine learning applied to medical imaging research community towards contributing to higher standards for developing reliable machine learning technology for medical applications.
Dr. Levman says the collaboration with Dr. Tyrrell, the Director of Data Science and associate professor in the Department of Medical Imaging at the University of Toronto, came about after they were introduced by Dr. Levman’s former PhD supervisor, Dr. Anne Martel.
“Dr. Tyrrell and my research interests overlap heavily, with both of us taking a keen interest in methods to establish reliable and reproducible machine learning technology for advanced applications in medical imaging. We have founded the Real-mi initiative (https://www.real-mi.org/), an effort towards establishing high standards in reproducibility and explainability in machine learning applied to medical imaging.”
