Adaptive Weighting Depth-variant Deconvolution of Fluorescence Microscopy Images with Convolutional Neural Network

• DOI: 10.48550/arxiv.1907.03217
• Publication Date: 2019-01-01
• Authors: Da He; De Cai; Jiasheng Zhou; Jiajia Luo; Sung‐Liang Chen
• Keywords: Point spread function

Fluorescence microscopy plays an important role in biomedical research. The depth-variant point spread function (PSF) of a fluorescence microscope produces low-quality images especially the out-of-focus regions thick specimens. Traditional deconvolution to restore is usually insufficient since depth-invariant PSF assumed. This article aims at handling by learning-based and reducing artifacts. We propose adaptive weighting (AWDVD) with defocus level prediction convolutional neural network (DelpNet) images. Depth-variant PSFs image patches can be obtained DelpNet applied afterward deconvolution. AWDVD adopted for whole which patch-wise deconvolved appropriately cropped before achieves accuracy 98.2%, outperforms best-ever one using same dataset. Image 11 levels after are validated maximum improvement peak signal-to-noise ratio structural similarity index 6.6 dB 11%, respectively. eliminate patch boundary artifacts improve quality. proposed method accurately estimate effectively recover To our acknowledge, this first study PSF. Facing most common blurs microscopy, novel provides practical technology