Developing novel quantitative imaging analysis schemes based machine learning for cancer research

The computer-aided detection (CAD) scheme is a developing technology in the medical imaging field, and it attracted extensive research interest in recent years. In this dissertation, I investigated the feasibility of developing several new novel CAD schemes for different cancer research purposes. First, I investigated the feasibility of identifying a new quantitative imaging marker based on false-positives generated by a computer-aided detection (CAD) scheme to predict short-term breast cancer risk. For this study, an existing CAD scheme was applied “as is” to process each image. From CAD-generated results, some detection features were computed from each image. Two logistic regression models were then trained and tested using a leave-one-case-out cross-validation method to predict each testing case's likelihood of being positive in the next subsequent screening. This study demonstrated that CAD-generated false-positives contain valuable information to predict short-term breast cancer risk. Second, I identified and applied quantitative imaging features computed from ultrasound images of athymic nude mice to predict tumor response to treatment at an early stage. For this study, a CAD scheme was developed to perform tumor segmentation and image feature analysis. The study demonstrated the feasibility of extracting quantitative image features from the ultrasound images taken at an early treatment stage to predict tumor response to therapies. Last, I optimized a machine learning model for predicting peritoneal metastasis in gastric cancer. For this purpose, I have developed a CAD scheme to segment the tumor volume and extract quantitative image features automatically. Then, I reduced the dimensionality of features with a new method named random projection to optimize the model's performance. Finally, the gradient boosting machine model was applied along with a synthetic minority oversampling technique to predict peritoneal metastasis risk. Results suggested that the random projection method yielded promising results in improving the accuracy performance in peritoneal metastasis prediction. In summary, in my Ph.D. studies, I have investigated and tested several innovative approaches to develop different CAD schemes and identify quantitative imaging markers with high discriminatory power in various cancer research applications. Study results demonstrated the feasibility of applying CAD technology to several new application fields, which can help radiologists and gynecologists improve accuracy and consistency in disease diagnosis and prognosis assessment of using the medical image.