Deep learning for laser beam imprinting.

Methods of ablation imprints in solid targets are widely used to characterize focused X-ray laser beams due to a remarkable dynamic range and resolving power. A detailed description of intense beam profiles is especially important in high-energy-density physics aiming at nonlinear phenomena. Complex interaction experiments require an enormous number of imprints to be created under all desired conditions making the analysis demanding and requiring a huge amount of human work. Here, for the first time, we present ablation imprinting methods assisted by deep learning approaches. Employing a multi-layer convolutional neural network (U-Net) trained on thousands of manually annotated ablation imprints in poly(methyl methacrylate), we characterize a focused beam of beamline FL24/FLASH2 at the Free-electron laser in Hamburg. The performance of the neural network is subject to a thorough benchmark test and comparison with experienced human analysts. Methods presented in this Paper pave the way towards a virtual analyst automatically processing experimental data from start to end.

Non-rigid consistent registration of 2D image sequences.

We present a novel algorithm for the registration of 2D image sequences that combines the principles of multiresolution B-spline-based elastic registration and those of bidirectional consistent registration. In our method, consecutive triples of images are iteratively registered to gradually extend the information through the set of images of the entire sequence. The intermediate results are reused for the registration of the following triple. We choose to interpolate the images and model the deformation fields using B-spline multiresolution pyramids. Novel boundary conditions are introduced to better characterize the deformations at the boundaries. In the experimental section, we quantitatively show that our method recovers from barrel/pincushion and fish-eye deformations with subpixel error. Moreover, it is more robust against outliers--occasional strong noise and large rotations--than the state-of-the-art methods. Finally, we show that our method can be used to realign series of histological serial sections, which are often heavily distorted due to folding and tearing of the tissues.

Parallel integral projection transform for straight electrode localization in 3-D ultrasound images.

In surgical practice, small metallic instruments are frequently used to perform various tasks inside the human body. We address the problem of their accurate localization in the tissue. Recent experiments using medical ultrasound have shown that this modality is suitable for real-time visualization of anatomical structures as well as the position of surgical instruments. We propose an image-processing algorithm that permits automatic estimation of the position of a line-segment-shaped object. This method was applied to the localization of a thin metallic electrode in biological tissue. We show that the electrode axis can be found through maximizing the parallel integral projection transform that is a form of the Radon transform. To accelerate this step, hierarchical mesh-grid algorithm is implemented. Once the axis position is known, localization of the electrode tip is performed. The method was tested on simulated images, on ultrasound images of a tissue mimicking phantom containing a metallic electrode, and on real ultrasound images from breast biopsy. The results indicate that the algorithm is robust with respect to variations in electrode position and speckle noise. Localization accuracy is of the order of hundreds of micrometers and is comparable to the ultrasound system axial resolution.

Unwarping of unidirectionally distorted EPI images.

Echo-planar imaging (EPI) is a fast nuclear magnetic resonance imaging (MRI) method. Unfortunately, local magnetic field inhomogeneities induced mainly by the subject's presence cause significant geometrical distortion, predominantly along the phase-encoding direction, which must be undone to allow for meaningful further processing. So far, this aspect has been too often neglected. In this paper, we suggest a new approach using an algorithm specifically developed for the automatic registration of distorted EPI images with corresponding anatomically correct MRI images. We model the deformation field with splines, which gives us a great deal of flexibility, while comprising the affine transform as a special case. The registration criterion is least squares. Interestingly, the complexity of its evaluation does not depend on the resolution of the control grid. The spline model gives us good accuracy thanks to its high approximation order. The short support of splines leads to a fast algorithm. A multiresolution approach yields robustness and additional speedup. The algorithm was tested on real as well as synthetic data, and the results were compared with a manual method. A wavelet-based Sobolev-type random deformation generator was developed for testing purposes. A blind test indicates that the proposed automatic method is faster, more reliable, and more precise than the manual one.