A Data-Centric Augmentation Approach for Disturbed Sensor Image Segmentation

Deep learning methods have become increasingly popular for optical sensor image analysis. They are adaptable to specific tasks and simultaneously demonstrate a high degree of generalization capability. However, applying deep neural networks to problems with low availability of labeled training data can lead to a model being incapable of generalizing to possible scenarios that may occur in test data, especially with the occurrence of dominant imaging artifacts. We propose a data-centric augmentation approach based on generative adversarial networks that overlays the existing labeled data with synthetic artifacts that are generated from data not present in the training set. This augmentation leads to a more robust generalization capability in semantic segmentation. Our method does not need any additional labeling and does not lead to additional memory or time consumption during inference. Further, we find it to be more effective than comparable approaches that are based on procedurally generated disturbances and the direct use of real disturbances. Building upon the improved segmentation results, we observe that our approach leads to improvements of 22% in the F1-score for an evaluated detection problem, which promises significant robustness towards future disturbances. In the context of sensor-based data analysis, the compensation of image artifacts is a challenge. When the structures of interest are not clearly visible in an image, algorithms that can cope with artifacts are crucial for obtaining the desired information. Thereby, the high variation of artifacts, the combination of different types of artifacts, and their similarity to signals of interest are specific issues that have to be considered in the analysis. Despite the high generalization capability of deep learning-based approaches, their recent success was driven by the availability of large amounts of labeled data. Therefore, the provision of comprehensive labeled image data with different characteristics of image artifacts is of importance. At the same time, applying deep neural networks to problems with low availability of labeled data remains a challenge. This work presents a data-centric augmentation approach based on generative adversarial networks that augments the existing labeled data with synthetic artifacts generated from data not present in the training set. In our experiments, this augmentation leads to a more robust generalization in segmentation. Our method does not need additional labeling and does not lead to additional memory or time consumption during inference. Further, we find it to be more effective than comparable augmentations based on procedurally generated artifacts and the direct use of real artifacts. Building upon the improved segmentation results, we observe that our approach leads to improvements of 22% in the F1-score for an evaluated detection problem. Having achieved these results with an example sensor, we expect increased robustness against artifacts in future applications.

Viewing and Lenses

How do displays—those of smartphones, tablets, laptops, televisions, and movie screens—fit with how people see? This chapter discusses acuity, visual fields, and how aspect ratios and screen size interact with them. It also considers lenses to photograph and project movie images, their lengths, their angles of view, and other effects such as their falloff in luminance and image artifacts that they create—bokeh and flare, depth of focus, and distortions of shape. Finally, it considers types of lenses—wide-angle, normal, and telephoto; spherical and anamorphic—and discusses how these are used by cinematographers to create emotional effects, the buildup of anxiety and the change of emotion, and how they can induce fearfulness.

Ultra-Homogeneous B0 field for High-Field Body Magnetic Resonance Imaging with Unified Shim-RF Coils

High-field magnetic resonance imaging (MRI, 3.0T and above) offers numerous advantages for imaging the human body over lower-field strengths. However, it suffers from unwanted fast spatially-varying main (B0 ) fields caused by the susceptibility mismatch at the tissue interfaces. When this is combined with the anatomical complexity of the human body, undesirable image artifacts can become damaging as they can compromise potential image contrasts, limit the use of accelerated imaging, and interfere with clinical interpretation. Consequently, these limitations restrict the effective utilization of high-field body MRI and emphasize the need for a major improvement in B0 field homogeneity to take full advantage of the ever increased B0 field. Here we introduce a Unified shim-RF Coil (UNIC) to overcome this existing bottleneck by transcending the conventional low-efficiency, distantly located B0 shim coils. UNIC allows a shim array to be freely allotted and seamlessly integrated into a standard surface RF coil, thus maximizing both the performance of RF receive sensitivity and effective B0 shimming. We demonstrate the capacity of the UNIC approach through detailed characterization of the coil design, prototyping a body coil integrating the UNIC features, and conducting in-vivo imaging of deep organs adjacent to the lung. Our studies provide evidence that UNIC enables homogeneous B0 fields in the liver and the heart, where strong image artifacts are known to occur, and hence facilitate the acquisition of unprecedented image quality in a clinical 3.0T scanner. Further, UNIC’s design is practical as it overcomes one of the most, if not the most, critical limitations of the state-of-the-art high-field MRI with minimal changes to the current MRI hardware architecture. Accordingly, the proposed technique offers opportunities for major advancements in noninvasive imaging of deep organs with high-field imaging in a way it has not been possible thus far.

PARAMETRIC BLIND-DECONVOLUTION METHOD TO REMOVE IMAGE ARTIFACTS IN WAVEFRONT CODING IMAGING SYSTEMS

Wavefront coding technique includes a phase mask of asymmetric phase mask kind in the pupil plane to extend the depth of field of an imaging system and the digital processing step to obtain the restored final high-quality image. However, the main drawback of wavefront coding technique is image artifacts on the restored final images. In this paper, we proposed a parameter blind-deconvolution method based on maximizing of the variance of the histogram of restored final images that enables to obtain the restored final image with artifact-free over a large range of defocus.

Fully Leveraging Deep Learning Methods for Constructing Retinal Fundus Photomontages

Retinal photomontages, which are constructed by aligning and integrating multiple fundus images, are useful in diagnosing retinal diseases affecting peripheral retina. We present a novel framework for constructing retinal photomontages that fully leverage recent deep learning methods. Deep learning based object detection is used to define the order of image registration and blending. Deep learning based vessel segmentation is used to enhance image texture to improve registration performance within a two step image registration framework comprising rigid and non-rigid registration. Experimental evaluation demonstrates the robustness of our montage construction method with an increased amount of successfully integrated images as well as reduction of image artifacts.

Silicon Carbide and MRI: Towards Developing a MRI Safe Neural Interface

An essential method to investigate neuromodulation effects of an invasive neural interface (INI) is magnetic resonance imaging (MRI). Presently, MRI imaging of patients with neural implants is highly restricted in high field MRI (e.g., 3 T and higher) due to patient safety concerns. This results in lower resolution MRI images and, consequently, degrades the efficacy of MRI imaging for diagnostic purposes in these patients. Cubic silicon carbide (3C-SiC) is a biocompatible wide-band-gap semiconductor with a high thermal conductivity and magnetic susceptibility compatible with brain tissue. It also has modifiable electrical conductivity through doping level control. These properties can improve the MRI compliance of 3C-SiC INIs, specifically in high field MRI scanning. In this work, the MRI compliance of epitaxial SiC films grown on various Si wafers, used to implement a monolithic neural implant (all-SiC), was studied. Via finite element method (FEM) and Fourier-based simulations, the specific absorption rate (SAR), induced heating, and image artifacts caused by the portion of the implant within a brain tissue phantom located in a 7 T small animal MRI machine were estimated and measured. The specific goal was to compare implant materials; thus, the effect of leads outside the tissue was not considered. The results of the simulations were validated via phantom experiments in the same 7 T MRI system. The simulation and experimental results revealed that free-standing 3C-SiC films had little to no image artifacts compared to silicon and platinum reference materials inside the MRI at 7 T. In addition, FEM simulations predicted an ~30% SAR reduction for 3C-SiC compared to Pt. These initial simulations and experiments indicate an all-SiC INI may effectively reduce MRI induced heating and image artifacts in high field MRI. In order to evaluate the MRI safety of a closed-loop, fully functional all-SiC INI as per ISO/TS 10974:2018 standard, additional research and development is being conducted and will be reported at a later date.