No abstract

The application of machine learning in wavefront reconstruction has brought great benefits to real-time, non-invasive, deep tissue imaging in biomedical research. However, due to the diversity and heterogeneity of biological tissues, it is difficult to train the dataset with a unified model. In general, the utilization of some unified models will result in the specific sample falling outside the training set, leading to low accuracy of the machine learning model in some real applications. This paper proposes a sensorless wavefront reconstruction method based on transfer learning to overcome the domain shift introduced by the difference between the training set and the target test set. We build a weights-sharing two-stream convolutional neural network (CNN) framework for the prediction of Zernike coefficient, in which a large number of labeled randomly generated samples serve as the source-domain data and the unlabeled specific samples serve as the target-domain data at the same time. By training on massive labeled simulated data with domain adaptation to unlabeled target-domain data, the network shows better performance on the target tissue samples. Experimental results show that the accuracy of the proposed method is 18.5% higher than that of conventional CNN-based method and the peak intensities of the point spread function (PSF) are more than 20% higher with almost the same training time and processing time. The better compensation performance on target sample could have more advantages when handling complex aberrations, especially the aberrations caused by various histological characteristics, such as refractive index inhomogeneity and biological motion in biological tissues.

Adaptive optics has been widely used in the optical microscopy to recover high-resolution images deep into the sample. However, the corrected field of view (FOV) with a single correction is generally limited, which seriously restricts the imaging speed. In this article, we demonstrate a high-speed wavefront correction method by using the conjugate adaptive optical correction with multiple guide stars (CAOMG) based on the coherent optical adaptive technique. The results show that the CAOMG method can greatly improve the corrected FOV. For 120-μm-thick mouse brain tissue, the corrected FOV can be improved up to ~243 times of the conventional pupil adaptive optics (PAO) without additional time consumption. Therefore, this study shows the potential of high-speed imaging through scattering medium in biological science.

Structured illumination microscopy (SIM) is widely used in biological imaging for its high resolution, fast imaging speed, and simple optical setup. However, when imaging thick samples, the structured illumination patterns in SIM will suffer from optical aberrations, leading to a serious deterioration in resolution. Therefore, it is necessary to reconstruct structured illumination patterns with high quality and efficiency in deep tissue imaging. Here we demonstrate an adaptive optics (AO) correction method based on deep learning in wide-field SIM imaging system. The mapping between the coefficients of the first 15 Zernike modes and their corresponding distorted patterns is established to train the convolution neural network (CNN). The results show that the optimized CNN can predict the aberration phase within 10.1 ms with a personal computer. The correlation index between the aberration phases and their corresponding predicted aberration phase is up to 0.9986. This method is highly robust and effective for patterns with various spatial densities and illumination conditions and able to effectively correct the imaging distortion caused by optical aberration in SIM system.

The doughnut beam is a spatially structured beam which has been widely used in super-resolution microscopy, laser trapping and so on. However, when it passes through thick scattering medium, aberrations will seriously affect its performance. Currently, adaptive optics (AO) has become one of the most powerful tools to compensate aberrations. However, conventional AO always suffers from limited corrected field of view (FOV). Here, we propose a method with conjugate AO system based on coherent optical adaptive technique. The results show that the corrected FOV can be improved effectively. For a wide range of the optical applications with doughnut beam, our method has potentials in correcting aberrations with high speed in turbid media.(A) Mouse brain slice, (B) the distribution of r PAO , (C) the distribution of r CAO . The vortex beam focus of the blue point in (B) and (C) among a 137.5 × 137.5 μm FOV (D1) ideally, (D2) with scattering, (D3) in pupil AO system and (D4) in conjugate AO system. K E Y W O R D S aberration correction, adaptive optics, doughnut beam, turbid medium