Applying novel machine learning technology to optimize computer-aided detection and diagnosis of medical images

The purpose of developing Computer-Aided Detection (CAD) schemes is to assist physicians (i.e., radiologists) in interpreting medical imaging findings and reducing inter-reader variability more accurately. In developing CAD schemes, Machine Learning (ML) plays an essential role because it is widely used to identify effective image features from complex datasets and optimally integrate them with the classifiers, which aims to assist the clinicians to more accurately detect early disease, classify disease types and predict disease treatment outcome. In my dissertation, in different studies, I assess the feasibility of developing several novel CAD systems in the area of medical imaging for different purposes. The first study aims to develop and evaluate a new computer-aided diagnosis (CADx) scheme based on analysis of global mammographic image features to predict the likelihood of cases being malignant. CADx scheme is applied to pre-process mammograms, generate two image maps in the frequency domain using discrete cosine transform and fast Fourier transform, compute bilateral image feature differences from left and right breasts, and apply a support vector machine (SVM) method to predict the likelihood of the case being malignant. This study demonstrates the feasibility of developing a new global image feature analysis based CADx scheme of mammograms with high performance. This new CADx approach is more efficient in development and potentially more robust in future applications by avoiding difficulty and possible errors in breast lesion segmentation. In the second study, to automatically identify a set of effective mammographic image features and build an optimal breast cancer risk stratification model, I investigate advantages of applying a machine learning approach embedded with a locally preserving projection (LPP) based feature combination and regeneration algorithm to predict short-term breast cancer risk. To this purpose, a computer-aided image processing scheme is applied to segment fibro-glandular tissue depicted on mammograms and initially compute 44 features related to the bilateral asymmetry of mammographic tissue density distribution between left and right breasts. Next, an embedded LLP algorithm optimizes the feature space and regenerates a new operational vector with 4 features using a maximal variance approach. This study demonstrates that applying the LPP algorithm effectively reduces feature dimensionality, and yields higher and potentially more robust performance in predicting short-term breast cancer risk. In the third study, to more precisely classify malignant lesions, I investigate the feasibility of applying a random projection algorithm to build an optimal feature vector from the initially CAD-generated large feature pool and improve the performance of the machine learning model. In this process, a CAD scheme is first applied to segment mass regions and initially compute 181 features. An SVM model embedded with the feature dimensionality reduction method is then built to predict the likelihood of…