Deep 3D reconstruction of synchrotron X-ray computed tomography for intact lungs.

Synchrotron X-rays can be used to obtain highly detailed images of parts of the lung. However, micro-motion artifacts induced by such as cardiac motion impede quantitative visualization of the alveoli in the lungs. This paper proposes a method that applies a neural network for synchrotron X-ray Computed Tomography (CT) data to reconstruct the high-quality 3D structure of alveoli in intact mouse lungs at expiration, without needing ground-truth data. Our approach reconstructs the spatial sequence of CT images by using a deep-image prior with interpolated input latent variables, and in this way significantly enhances the images of alveolar structure compared with the prior art. The approach successfully visualizes 3D alveolar units of intact mouse lungs at expiration and enables us to measure the diameter of the alveoli. We believe that our approach helps to accurately visualize other living organs hampered by micro-motion.

Deep Learning Recurrent Neural Network for Concussion Classification in Adolescents Using Raw Electroencephalography Signals: Toward a Minimal Number of Sensors.

Artificial neural networks (ANNs) are showing increasing promise as decision support tools in medicine and particularly in neuroscience and neuroimaging. Recently, there has been increasing work on using neural networks to classify individuals with concussion using electroencephalography (EEG) data. However, to date the need for research grade equipment has limited the applications to clinical environments. We recently developed a deep learning long short-term memory (LSTM) based recurrent neural network to classify concussion using raw, resting state data using 64 EEG channels and achieved high accuracy in classifying concussion. Here, we report on our efforts to develop a clinically practical system using a minimal subset of EEG sensors. EEG data from 23 athletes who had suffered a sport-related concussion and 35 non-concussed, control athletes were used for this study. We tested and ranked each of the original 64 channels based on its contribution toward the concussion classification performed by the original LSTM network. The top scoring channels were used to train and test a network with the same architecture as the previously trained network. We found that with only six of the top scoring channels the classifier identified concussions with an accuracy of 94%. These results show that it is possible to classify concussion using raw, resting state data from a small number of EEG sensors, constituting a first step toward developing portable, easy to use EEG systems that can be used in a clinical setting.

Eigendecomposition-Free Training of Deep Networks for Linear Least-Square Problems.

Many classical Computer Vision problems, such as essential matrix computation and pose estimation from 3D to 2D correspondences, can be tackled by solving a linear least-square problem, which can be done by finding the eigenvector corresponding to the smallest, or zero, eigenvalue of a matrix representing a linear system. Incorporating this in deep learning frameworks would allow us to explicitly encode known notions of geometry, instead of having the network implicitly learn them from data. However, performing eigendecomposition within a network requires the ability to differentiate this operation. While theoretically doable, this introduces numerical instability in the optimization process in practice. In this paper, we introduce an eigendecomposition-free approach to training a deep network whose loss depends on the eigenvector corresponding to a zero eigenvalue of a matrix predicted by the network. We demonstrate that our approach is much more robust than explicit differentiation of the eigendecomposition using two general tasks, outlier rejection and denoising, with several practical examples including wide-baseline stereo, the perspective-n-point problem, and ellipse fitting. Empirically, our method has better convergence properties and yields state-of-the-art results.

Technical Note: Synthesizing of lung tumors in computed tomography images.

PURPOSE: When investigating new radiation therapy techniques in the treatment planning stage, it can be extremely time consuming to locate multiple patient scans that match the desired characteristics for the treatment. With the help of machine learning, we propose to bypass the difficulty in finding patient computed tomography (CT) scans that match the treatment requirements. Furthermore, we aim to provide the developed method as a tool that is easily accessible to interested researchers. METHODS: We propose a generative adversarial network (GAN) to edit individual volumes of interest (VOIs) in pre-existing CT scans, translating features of the healthy VOIs into features of cancerous volumes. Training and testing was done using VOIs from a dataset of 460 diagnostic and lung cancer screening CT scans. Agreement between real tumors and those produced by the editor was tested by comparing the distributions of several histogram parameters and second-order statistics as well as using qualitative analysis. RESULTS: After training, the network was successfully able to map healthy CT segments to realistic looking cancerous volumes. Based on visual inspection, tumors produced by the editor were found to be both realistic and visually consistent with the surrounding anatomy when placed back into the original CT scan. Furthermore, the network was found to be able to extrapolate well beyond the upper size limit of the training set. Lastly, a graphical user interface (GUI) was developed to easily interact with the resulting network. CONCLUSION: The trained network and associated GUI can serve as a tool to develop an abundance of lung cancer patient data to be used in treatment planning. In addition, this method can be extended to a variety of cancer types if given an appropriate baseline dataset. The GUI and instructions on how to utilize the tool have been made publicly available at https://github.com/teaghan/CT_Editor.

Robust 3D Object Tracking from Monocular Images Using Stable Parts.

We present an algorithm for estimating the pose of a rigid object in real-time under challenging conditions. Our method effectively handles poorly textured objects in cluttered, changing environments, even when their appearance is corrupted by large occlusions, and it relies on grayscale images to handle metallic environments on which depth cameras would fail. As a result, our method is suitable for practical Augmented Reality applications including industrial environments. At the core of our approach is a novel representation for the 3D pose of object parts: We predict the 3D pose of each part in the form of the 2D projections of a few control points. The advantages of this representation is three-fold: We can predict the 3D pose of the object even when only one part is visible; when several parts are visible, we can easily combine them to compute a better pose of the object; the 3D pose we obtain is usually very accurate, even when only few parts are visible. We show how to use this representation in a robust 3D tracking framework. In addition to extensive comparisons with the state-of-the-art, we demonstrate our method on a practical Augmented Reality application for maintenance assistance in the ATLAS particle detector at CERN.