Analysis of Optical Coherence Tomography Images by Dictionary Learning Methods

von Raha Razavi
Dissertation
Datum der mündl. Prüfung:2024-07-08
Erschienen:2024-07-17
Betreuer:Prof. Dr. Gerlind Plonka-Hoch
Gutachter:Prof. Dr. Gerlind Plonka-Hoch
Gutachter:Prof. Dr. Luke Russell
Zum Verlinken/Zitieren: http://dx.doi.org/10.53846/goediss-10616

 

 

Dateien

Name:Thesis-Raha Razavi.pdf
Size:68.6Mb
Format:PDF

Lizenzbestimmungen:

Zusammenfassung

Englisch

An accurate analysis of Optical Coherence Tomography (OCT) images plays an important role in the diagnosis of abnormalities within these images. Unfortunately, noise caused by various sources degenerates the quality of the OCT images. Among all, OCT images suffer mainly from speckle noise caused by the scattering of light waves in the physical device. In this thesis, we propose new denoising methods for OCT images and investigate the noise removal impact on segmentation and classification results. We consider sparse representation modeling and aim to benefit from non-data-adaptive multi-scale ($X$-let) transforms. We study these transforms and compare their performances for OCT images. These transforms, besides their stable performance, are also useful for the analysis of different types of OCT images. We employ and adapt decomposition algorithms for OCT images with noise reduction applications within the transform domain of $X$-let transforms. The obtained components from these methods are distinct in texture, piecewise smooth parts, and singularities along curves. Numerically and visually, we achieve strongly enhanced quality and denoised OCT images by applying adaptive local thresholding techniques separately to each image component. The denoising performance outperforms other state-of-the-art denoising algorithms regarding the different image quality measures. Lastly, we emphasize and demonstrate the significance of denoising as the preprocessing step for segmentation and classification results. The acquired decomposition of OCT images into well-interpretable (denoised) image components using our proposed algorithms can be exploited further for image processing tasks.
Keywords: Optical Coherence Tomography; OCT; Denoising; Image Processing; Medical Image Processing; Non-data-adaptive Transforms
 
Statistik