The rapid growth of diagnostic medical imaging studies has led to enormous strides in the effective diagnosis and treatment of a myriad of diseases, from chronic diseases to life threatening cancers. The rise of imaging as a major factor in medical decision making has directly led to a drive towards quantification of image findings, in order to augment the qualitative analysis of trained medical professionals, such as radiologists. The overarching goal of this dissertation is to explore, develop, and evaluate imaging biomarkers for both chronic and life threatening diseases. Towards this goal, this dissertation has the following two aims:
1. Discover and characterize imaging biomarkers for diseases which may have either acute/sub-acute presentation or treatment, or diseases which may have a more chronic course and treatment intervention. The former analysis is focused on the case of cervical cancer, and the latter is focused on osteoarthritis (OA).
2. Apply digital image processing methods to the acquired images which leverage the spatial information in the images to characterize the anatomy, physiology, and pathophysiology of a disease process.
These aims were designed to improve the characterization of human disease by extracting quantifiable information from the acquired medical images. For the case of cervical cancer, analyses were undertaken of higher order statistics and texture measures as features to be used in the classification of cervical cancer tumor volumes in dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) as either likely to recur or be controlled after treatment. The contributions of these analyses are:
1. A physiologically meaningful measure of tumor vascularity by using the wavelet decomposition to quantify texture, which predicts the outcome of radiation treatment at two-year post-treatment follow-up with 100% accuracy, based on early treatment DCE-MRI studies of the tumor.
2. The demonstration of change in texture measures as more important than change in tumor volume for accurate treatment outcome prediction.
For the case of OA, analyses were undertaken of the quadriceps muscle morphology (cross-sectional area (CSA)) and content (intramuscular adipose tissue (IAT)), femur morphology, and meniscus morphology (volume) on MRI as potential biomarkers for OA severity. The analyses of the quadriceps muscle and the femur were further pursued through the development of semi-automated and automated segmentation procedures. The findings and contributions of these analyses are:
1. The association of the CSA of a particular muscle in the quadriceps, the vastus intermedius, with a reduced risk of more severe radiographic OA.
2. The finding that quadriceps lean muscle CSA (anatomical muscle CSA minus IAT) is associated with sex, age, and body mass index (BMI), which are risk factors for OA.
3. The volume of the lateral meniscus has multiple significant associations with the volume of the tibiofemoral articular cartilage in subjects with OA.
4. The development of an image enhancement procedure which standardizes the intensities in MRI images between all subjects and allows for efficient and accurate structure segmentation and imaging biomarker calculation.
5. The development of a semi-automated segmentation algorithm for the individual quadriceps muscles, using atlases and level set contour evolution.
6. A comparison of atlas segmentation procedures, demonstrating that the use of multiple atlases is superior to the use of “representative” atlases selected by a trained human reader.
The use of medical imaging allows for the non-invasive localization and characterization of disease, which is especially important for diseases which have significant mortality (such as cervical cancer) or long-term morbidity (such as OA). By studying diseases at these two ends of the spectrum, a unique understanding of the imaging biomarker development and evaluation process can be gained. In addition, the application of digital image processing methods can improve the characterization of the imaging manifestations of disease by providing consistent, accurate, and complicated quantification techniques, which may be difficult, or impossible, for human readers to provide.