Hide/Show Apps

Single-image bayesian restoration and multi-image super-resolution restoration for b-mode ultrasound images using an accurate system model

Cüneyitoğlu Özkul, Min
Medical imaging is an essential part of diagnosis and intervention/surgery planning in modern medicine. Compared to other medical imaging modalities, ultrasound provides a variety of diagnostic advantages. The imaging equipment is less expensive and more portable. There is no known harm to human tissue, therefore, it is applicable in almost any medical field safely. However, ultrasound image quality is usually poorer compared to other modalities. If the image quality of medical ultrasound is improved, it will be beneficial for clinical usage. This makes the research on this subject an increasingly important field, both academically and commercially. In this thesis, image quality improvement was aimed. Both single and multi-frame, in-plane, freehand, 2D, B-mode ultrasound scan data was used for this purpose. Non-rigid image registration, Bayesian image restoration and super-resolution methods, along with a detailed study on statistical modelling of the speckle was employed. Tissue-mimicking resolution phantoms were used to characterize the clinical imaging system. The methods were then tested on a tissue-mimicking breast phantom and on various superficial tissue images collected from volunteers. The methods developed were compared to the well-known image restoration and filtering methods in the literature. Relevant objective image quality evaluation metrics were used to measure overall improvements. Additionally, expert opinions were obtained and evaluated using visual grading analysis. The proposed methods have the drawback of increased computation time. However, the prominent contribution of this study is the improvement in image quality according to the results of objective evaluation and opinions of experts. If high-end image processing hardware is used to reduce implementation time, the proposed methods may be useful for clinical applications.