Deep learning based noise reduction method for automatic 3Dsegmentation of the anterior of lamina cribrosa in optical coherencetomography volumetric scans

摘要

A deep-learning (DL) based noise reduction algorithm, in combinationwith a vessel shadow compensation method and a three-dimensional (3D)segmentation technique, has been developed to achieve, to the authorsbest knowledge, the first automatic segmentation of the anteriorsurface of the lamina cribrosa (LC) in volumetric ophthalmic opticalcoherence tomography (OCT) scans. The present DL-based OCT noisereduction algorithm was trained without the need of noise-free groundtruth images by utilizing the latest development in deep learning ofde-noising from single noisy images, and was demonstrated to be ableto cover more locations in the retina and disease cases of differenttypes to achieve high robustness. Compared with the original singleOCT images, a 6.6 dB improvement in peak signal-to-noise ratio and a0.65 improvement in the structural similarity index were achieved. Thevessel shadow compensation method analyzes the energy profile in eachA-line and automatically compensates the pixel intensity of locationsunderneath the detected blood vessel. Combining the noise reductionalgorithm and the shadow compensation and contrast enhancementtechnique, medical experts were able to identify the anterior surfaceof the LC in 98.3% of the OCT images. The 3D segmentationalgorithm employs a two-round procedure based on gradients informationand information from neighboring images. An accuracy of 90.6%was achieved in a validation study involving 180 individual B-scansfrom 36 subjects, compared to 64.4% in raw images. This imagingand analysis strategy enables the first automatic complete view of theanterior LC surface, to the authors best knowledge, which may have thepotentials in new LC parameters development for glaucoma diagnosis andmanagement.