Grant: Master's Award: Canada Graduate Scholarships

Radiomics is an image analysis technique that extracts numerical features from medical images, for example computed tomography (CT) images. These radiomic features can be used with machine learning models to predict outcomes. Machine learning models include two main steps: feature selection methods and classification methods. Chronic obstructive pulmonary disease (COPD) is an inflammatory lung disease that is caused by the interactions between genes and the environment. The current method for diagnosing COPD is with spirometry, which measures lung function. There are several limitations with spirometry that can lead to undiagnosed COPD. To overcome these limitations, the use of radiomic features can be used with machine learning to predict the presence of COPD. Although, with a large variety of different feature selection methods and classification methods, it is unknown what models have the highest prediction performance for COPD with radiomic features. The objective of this research is to compare the various combinations of feature selection and classification methods to determine the model with the highest accuracy for predicting COPD. Combinations of 17 feature selection and 8 classification methods will be tested and evaluated. The removal of outliers and the number of features selected for the classification model will also be tested to potentially improve the prediction accuracy of the model. The expected outcome from this research is to establish an optimal machine learning model for predicting COPD using extracted CT radiomic features. In addition, the optimal machine learning model could be applied to other lung diseases.