OCT images from different machines or different manufacturers exhibit very different noise characteristics, so it is challenging to develop a deep-learning-based algorithm to well remove the noise in OCT images. Deep-learning-based algorithms often over-fit to training data and have poor generalization performance, e,g, if training and testing images are obtained from different imaging sources, the denoising results would degrade significantly. In recent years, a lot of denoising algorithms for OCT images have been developed to improve the image quality, which can be broadly divided into three categories: deep-learning-based 13, 14, sparse-coding-based 15, 16, 17, 18 and filter-based 19, 20, 21. In current commercial retinal OCT machines, a common optical approach to improve the quality of OCT images is to conduct averaging through overlapping scans 9, 10 or incoherent averaging 11, 12, but it has two disadvantages: (1) image quality degradation due to eye movement, which often happens in a clinical settings, (2) extending the acquisition time so that it may take too long for OCT imaging. It is hence essential to enhance the medical image quality by removing noise from OCT images, and preserving image micro-structures such as edges at the same time.
A large amount of noise will also reduce the positioning accuracy of surgical instruments in intraoperative retinal OCT imaging. In medical diagnosis and therapy, besides affecting the overall image clarity, image noise and low contrast of images will also affect the image segmentation of lesion areas, which is critical for medical diagnosis and treatment. The visualization quality of retinal OCT is often degraded by noises from different sources such as limited light bandwidth, phase aberrations of propagating beam, the aperture of the detector and multiple scatters within the coherence length 8. The noise pattern will change if the surface moves slightly 7. The dark and bright spots formed by the reflected beam have no obvious relationship with the surface texture, which are often seen as noise. OCT depends fundamentally on the coherence of the light used in the imaging process, and hence the reflection of a laser beam from a rough surface has a distinctive granular or mottle appearance 6. Moreover, the image-acquisition speed of OCT systems has been enhanced along with the development of high-speed sensors and tunable lasers with MHz scanning rate, to facilitate real-time retinal imaging. Fourier-domain retinal OCT 4, 5 is able to image biological tissues at a cellular level, and up to the depth of 1 mm below the retinal surface with high image quality. It has been widely used particularly in ophthalmology and other fields 3 because of its non-invasive nature and high resolution. The resolutions are in the range of 1 to 15 µm, smaller than that of ultrasound imaging 2. Optical Coherence Tomography (OCT) 1 is an established medical imaging technique that captures micrometer-resolution, three-dimensional images by imposing light on optical scattering media such as biological tissue. The comparison experimental results on three benchmark datasets against 8 compared algorithms demonstrate the effectiveness of the proposed approach in removing OCT noise. The proposed adaptive thresholding scheme optimally adapts to various noise conditions and hence better remove the noise. The proposed algorithm effectively transforms the Poisson noise to Gaussian noise so that the subsequent shearlet transform could optimally remove the noise.
Compared with other algorithms, the proposed algorithm better removes noise in OCT images and better preserves image details, significantly outperforming in terms of both quantitative evaluation and visual inspection. The proposed adaptive-SIN is evaluated on three benchmark datasets using quantitative evaluation metrics and subjective visual inspection. We propose an adaptive 3D shearlet image filter with noise-redistribution (adaptive-SIN) scheme for OCT images. Different manufacturers and differences between imaging objects may influence the observed noise characteristics, which make predefined thresholding scheme ineffective.
We hence propose a square-root transform to redistribute the OCT noise. The OCT noise is closer to the Poisson distribution than the Gaussian distribution, and shearlet transform assumes additive white Gaussian noise. In the paper, we propose an adaptive denoising algorithm for OCT images. Shearlet transform has shown its effectiveness in removing image noises because of its edge-preserving property and directional sensitivity. Optical coherence tomography (OCT) images is widely used in ophthalmic examination, but their qualities are often affected by noises.
0 kommentar(er)
