Medical Imaging Networks Research Articles

Deep learning for mass detection in Full Field Digital Mammograms

In recent years, the use of Convolutional Neural Networks (CNNs) in medical imaging has shown improved performance in terms of mass detection and classification compared to current state-of-the-art methods. This paper proposes a fully automated framework to detect masses in Full-Field Digital Mammograms (FFDM). This is based on the Faster Region-based Convolutional Neural Network (Faster-RCNN) model and is applied for detecting masses in the large-scale OPTIMAM Mammography Image Database (OMI-DB), which consists of ∼80,000 FFDMs mainly from Hologic and General Electric (GE) scanners. This research is the first to benchmark the performance of deep learning on OMI-DB. The proposed framework obtained a True Positive Rate (TPR) of 0.93 at 0.78 False Positive per Image (FPI) on FFDMs from the Hologic scanner. Transfer learning is then used in the Faster R-CNN model trained on Hologic images to detect masses in smaller databases containing FFDMs from the GE scanner and another public dataset INbreast (Siemens scanner). The detection framework obtained a TPR of 0.91±0.06 at 1.69 FPI for images from the GE scanner and also showed higher performance compared to state-of-the-art methods on the INbreast dataset, obtaining a TPR of 0.99±0.03 at 1.17 FPI for malignant and 0.85±0.08 at 1.0 FPI for benign masses, showing the potential to be used as part of an advanced CAD system for breast cancer screening.

Overview of image-to-image translation by use of deep neural networks: Denoising, super-resolution, modality conversion, and reconstruction in medical imaging

Since the advent of deep convolutional neural networks (DNNs), computer vision has seen an extremely rapid progress that has led to huge advances in medical imaging. Every year, many new methods are reported at conferences such as the International Conference on Medical Image Computing and Computer-Assisted Intervention and Machine Learning for Medical Image Reconstruction, or published online at the preprint server arXiv. There is a plethora of surveys on applications of neural networks in medical imaging (see [1] for a relatively recent comprehensive survey). This article does not aim to cover all aspects of the field, but focuses on a particular topic, image-to-image translation. Although the topic may not sound familiar, it turns out that many seemingly irrelevant applications can be understood as instances of image-to-image translation. Such applications include (1) noise reduction, (2) super-resolution, (3) image synthesis, and (4) reconstruction. The same underlying principles and algorithms work for various tasks. Our aim is to introduce some of the key ideas on this topic from a uniform viewpoint. We introduce core ideas and jargon that are specific to image processing by use of DNNs. Having an intuitive grasp of the core ideas of applications of neural networks in medical imaging and a knowledge of technical terms would be of great help to the reader for understanding the existing and future applications. Most of the recent applications which build on image-to-image translation are based on one of two fundamental architectures, called pix2pix and CycleGAN, depending on whether the available training data are paired or unpaired (see Sect. 1.3). We provide codes ([2, 3]) which implement these two architectures with various enhancements. Our codes are available online with use of the very permissive MIT license. We provide a hands-on tutorial for training a model for denoising based on our codes (see Sect. 6). We hope that this article, together with the codes, will provide both an overview and the details of the key algorithms and that it will serve as a basis for the development of new applications.

Overview of image-to-image translation by use of deep neural networks: denoising, super-resolution, modality conversion, and reconstruction in medical imaging.

Since the advent of deep convolutional neural networks (DNNs), computer vision has seen an extremely rapid progress that has led to huge advances in medical imaging. Every year, many new methods are reported at conferences such as the International Conference on Medical Image Computing and Computer-Assisted Intervention and Machine Learning for Medical Image Reconstruction, or published online at the preprint server arXiv. There is a plethora of surveys on applications of neural networks in medical imaging (see [1] for a relatively recent comprehensive survey). This article does not aim to cover all aspects of the field, but focuses on a particular topic, image-to-image translation. Although the topic may not sound familiar, it turns out that many seemingly irrelevant applications can be understood as instances of image-to-image translation. Such applications include (1) noise reduction, (2) super-resolution, (3) image synthesis, and (4) reconstruction. The same underlying principles and algorithms work for various tasks. Our aim is to introduce some of the key ideas on this topic from a uniform viewpoint. We introduce core ideas and jargon that are specific to image processing by use of DNNs. Having an intuitive grasp of the core ideas of applications of neural networks in medical imagingand a knowledge of technical terms would be of great help to the reader for understanding the existing and future applications. Most of the recent applications which build on image-to-image translation are based on one of two fundamental architectures, called pix2pix and CycleGAN, depending on whether the available training data are paired or unpaired (see Sect.1.3). We provide codes ([2, 3]) which implement these two architectures with various enhancements. Our codes are available online with use of the very permissive MIT license. We provide a hands-on tutorial for training a model for denoising based on our codes (see Sect.6). We hope that this article, together with the codes, will provide both an overview and the details of the key algorithms and that it will serve as a basis for the development of new applications.

Automatic mass detection in mammograms using deep convolutional neural networks.

With recent advances in the field of deep learning, the use of convolutional neural networks (CNNs) in medical imaging has become very encouraging. The aim of our paper is to propose a patch-based CNN method for automated mass detection in full-field digital mammograms (FFDM). In addition to evaluating CNNs pretrained with the ImageNet dataset, we investigate the use of transfer learning for a particular domain adaptation. First, the CNN is trained using a large public database of digitized mammograms (CBIS-DDSM dataset), and then the model is transferred and tested onto the smaller database of digital mammograms (INbreast dataset). We evaluate three widely used CNNs (VGG16, ResNet50, InceptionV3) and show that the InceptionV3 obtains the best performance for classifying the mass and nonmass breast region for CBIS-DDSM. We further show the benefit of domain adaptation between the CBIS-DDSM (digitized) and INbreast (digital) datasets using the InceptionV3 CNN. Mass detection evaluation follows a fivefold cross-validation strategy using free-response operating characteristic curves. Results show that the transfer learning from CBIS-DDSM obtains a substantially higher performance with the best true positive rate (TPR) of at 1.67 false positives per image (FPI), compared with transfer learning from ImageNet with TPR of at 2.1 FPI. In addition, the proposed framework improves upon mass detection results described in the literature on the INbreast database, in terms of both TPR and FPI.

Randomized Projection Methods for Linear Systems with Arbitrarily Large Sparse Corruptions

In applications like medical imaging, error correction, and sensor networks, one needs to solve large-scale linear systems that may be corrupted by a small number of arbitrarily large corruptions. ...

Compressive Sensing of Noisy 3-D Images Based on Threshold Selection

Compressive sensing (CS) of high-order data such as hyperspectral images, medical imaging, video sequences, and multi-sensor networks is certainly a hot issue after the emergence of tensor decomposition. Actually, the reconstruction accuracy with current algorithms is not ideal in some cases of noise. In this paper, we propose a new method that can recover noisy 3-D images from a reduced set of compressive measurements. First, multi-way compressive measurements are performed using Gaussian random matrices. Second, the mapping relationship between the variance of noise and the reconstruction threshold is found. Finally, the original images are recovered through reconstruction of pseudo inverse based on threshold selection. We experimentally demonstrate that the proposed method outperforms other similar methods in both reconstruction accuracy (within a range of the compression ratios and different variances of noise) and processing speed.

Convolutional Neural Networks in Medical Imaging

Convolutional Neural Networks in Medical Imaging

Geometric Graph Matching Using Monte Carlo Tree Search

We present an efficient matching method for generalized geometric graphs. Such graphs consist of vertices in space connected by curves and can represent many real world structures such as road networks in remote sensing, or vessel networks in medical imaging. Graph matching can be used for very fast and possibly multimodal registration of images of these structures. We formulate the matching problem as asingle player game solved using Monte Carlo Tree Search, which automatically balances exploring new possible matches and extending existing matches. Our method can handle partial matches, topological differences, geometrical distortion, does not use appearance information and does not require an initial alignment. Moreover, our method is very efficient-it can match graphs with thousands of nodes, which is an order of magnitude better than the best competing method, and the matching only takes a few seconds.

Approaches to improving the network teaching of medical imaging

In view of the problems existing in medical image network teaching, such as redundant and rigid curriculum system, obsolete structure, untimely updates, lack of supervision system and poor stability of network security and so on, we explored how to improve the medical image network teaching from the fusion of traditional teaching, optimizing curriculum system, perfecting the medical imaging system of autonomous learning, encouraging online communication, creating a good atmosphere, increasing invest-ment in software development and multi-sectoral collaboration, in order to complete the transfer of knowl-edge more effectively, and promote the good development of medical imaging teaching. Key words: Medical imaging; Network-based teaching; Teaching mode

The impact of medical informatics on safety and quality assurance

Health Informatics have radically changed the way medical imaging and radiotherapy services are delivered and they continue to evolve rapidly. Apart from the modality-specific post-processing techniques that have improved the clinical outcomes of the procedures and led to the development of novel techniques, medical networks has also played an important role in improving the quality and safety during the delivery of those services. Widespread use of networks has allowed for more efficient sharing of important information about the patient, his medical history and the requirements of the requested procedures. This can improve the workflow, lead to more appropriate planning of the procedures and radically decrease the possible hazards and adverse effects on the patient. Besides, the availability of exam data and meta-data can provide valuable information which can be used to assess the established practices and point out possible shortcomings or opportunities for possible improvement. Medical imaging networks operate according to established standards that ensure the exchange of data between the components of the network and the modalities but allow certain flexibility in their implementation in order to be applicable in all different modalities and facilities. So, it is important responsibility of all the professionals involved in all steps of the work routine to refine the use of network data according to the needs of the facility and develop consistent and efficient ways of handling all possible challenges. Health Informatics have radically changed the way medical imaging and radiotherapy services are delivered and they continue to evolve rapidly. Apart from the modality-specific post-processing techniques that have improved the clinical outcomes of the procedures and led to the development of novel techniques, medical networks has also played an important role in improving the quality and safety during the delivery of those services. Widespread use of networks has allowed for more efficient sharing of important information about the patient, his medical history and the requirements of the requested procedures. This can improve the workflow, lead to more appropriate planning of the procedures and radically decrease the possible hazards and adverse effects on the patient. Besides, the availability of exam data and meta-data can provide valuable information which can be used to assess the established practices and point out possible shortcomings or opportunities for possible improvement. Medical imaging networks operate according to established standards that ensure the exchange of data between the components of the network and the modalities but allow certain flexibility in their implementation in order to be applicable in all different modalities and facilities. So, it is important responsibility of all the professionals involved in all steps of the work routine to refine the use of network data according to the needs of the facility and develop consistent and efficient ways of handling all possible challenges.

Adaptive tensor compressive sensing based on noise estimation: application in three-dimensional images

Tensor Compressive Sensing (CS) is an emerging approach for higher order data representation, such as medical imaging, video sequences and multi-sensor networks. In this paper, we propose an Adaptive Tensor CS (ATCS) scheme for Three-dimensional (3D) images, especially those which contain noise. First, we find the relationship between reconstruction performance, noise level and sampling rate. Second, we develop the ATCS method by implementing a noise estimation algorithm. Finally, we apply the method in the CS system for efficient representation of 3D video sequences. We also demonstrate experimentally that ATCS outperforms other state of the art algorithms.

Analyticity of Semisimple Eigenvalues and Corresponding Eigenvectors of Matrix-Valued Functions

In this paper, we study the analyticity of the semisimple eigenvalue and the corresponding eigenvector functions of general analytic matrix-valued functions which are very important in many engineering applications such as the optimum design of dynamical structures, model updating, damage detecting, quantum mechanics, diffraction grating theory, medical imaging, and social network theory. We establish some new sufficient conditions on existence of analytic eigenvalue and eigenvector functions corresponding to semisimple eigenvalues of general analytic matrix-valued functions, which relaxes the condition in [P. Lancaster, A. S. Markus, and F. Zhou, SIAM J. Matrix Anal. Appl., 25 (2003), pp. 606--626]. We also present a numerical method to compute the derivatives of the eigenvalue and eigenvector functions. Numerical performance of the method is illustrated by some numerical examples.

A Routing Mechanism for Cloud Outsourcing of Medical Imaging Repositories.

Web-based technologies have been increasingly used in picture archive and communication systems (PACS), in services related to storage, distribution, and visualization of medical images. Nowadays, many healthcare institutions are outsourcing their repositories to the cloud. However, managing communications between multiple geo-distributed locations is still challenging due to the complexity of dealing with huge volumes of data and bandwidth requirements. Moreover, standard methodologies still do not take full advantage of outsourced archives, namely because their integration with other in-house solutions is troublesome. In order to improve the performance of distributed medical imaging networks, a smart routing mechanism was developed. This includes an innovative cache system based on splitting and dynamic management of digital imaging and communications in medicine objects. The proposed solution was successfully deployed in a regional PACS archive. The results obtained proved that it is better than conventional approaches, as it reduces remote access latency and also the required cache storage space.

A framework for integration of heterogeneous medical imaging networks.

Medical imaging is increasing its importance in matters of medical diagnosis and in treatment support. Much is due to computers that have revolutionized medical imaging not only in acquisition process but also in the way it is visualized, stored, exchanged and managed. Picture Archiving and Communication Systems (PACS) is an example of how medical imaging takes advantage of computers. To solve problems of interoperability of PACS and medical imaging equipment, the Digital Imaging and Communications in Medicine (DICOM) standard was defined and widely implemented in current solutions. More recently, the need to exchange medical data between distinct institutions resulted in Integrating the Healthcare Enterprise (IHE) initiative that contains a content profile especially conceived for medical imaging exchange: Cross Enterprise Document Sharing for imaging (XDS-i). Moreover, due to application requirements, many solutions developed private networks to support their services. For instance, some applications support enhanced query and retrieve over DICOM objects metadata.This paper proposes anintegration framework to medical imaging networks that provides protocols interoperability and data federation services. It is an extensible plugin system that supports standard approaches (DICOM and XDS-I), but is also capable of supporting private protocols. The framework is being used in the Dicoogle Open Source PACS.

Axiomatics for oriented connectivity

Tankyevych et al. [17] considered, in a directed graph, any set S of vertices where there is some marker vertex p∈S such that for every vertex x∈S, there is a directed path from p to x included in S; the family of all such sets S was called a semi-connection. Their properties were briefly analysed and compared with connectivity and connected components in undirected graphs.We give an abstract algebraic formalization of this concept, following the same approach as that of Serra [14] and Ronse [9] for the notions of connection and partial connection, which generalize both topological and graph-theoretic connectivity. Here the sets S are unsufficient, one must associate to them their markers p; thus in a space E we consider a family R of ordered pairs (p,S)∈E×P(E), where the set S can be “reached” from marker p; this family, which we call a reach, must satisfy the three properties of union, transitivity and membership; a fourth point property leads to a full reach. As in [14,9], we give an equivalent definition in terms of a system of point openings (γp,p∈E) satisfying some properties. The special case of symmetry, where S does not depend on the choice of the marker p∈S, leads to a partial connection or a connection. Some examples are given.Possible applications of this new theory lie in the analysis of connected structures having an orientation, for instance vascular networks in medical imaging. One can also apply it to geodesic reconstruction and connected filtering.

High Performance Computational Systems Biology

This paper includes a selection of papers presented at the Second International Workshop on High Performance Computational Systems Biology (HiBi 2010), which was colocated with the joint ICGT/SPIN conference, that was held at the University of Twente, Enschede, The Netherlands, on 27-29 September 2010. The selected papers cover a broad range of bioinformatics topics, including medical imaging, Markov clustering, simulation, model checking, and gene networks.

Medical image analysis with artificial neural networks

Given that neural networks have been widely reported in the research community of medical imaging, we provide a focused literature survey on recent neural network developments in computer-aided diagnosis, medical image segmentation and edge detection towards visual content analysis, and medical image registration for its pre-processing and post-processing, with the aims of increasing awareness of how neural networks can be applied to these areas and to provide a foundation for further research and practical development. Representative techniques and algorithms are explained in detail to provide inspiring examples illustrating: (i) how a known neural network with fixed structure and training procedure could be applied to resolve a medical imaging problem; (ii) how medical images could be analysed, processed, and characterised by neural networks; and (iii) how neural networks could be expanded further to resolve problems relevant to medical imaging. In the concluding section, a highlight of comparisons among many neural network applications is included to provide a global view on computational intelligence with neural networks in medical imaging.

Assessment of neural networks training strategies for histomorphometric analysis of synchrotron radiation medical images

Micro-computed tomography (μCT) obtained by synchrotron radiation (SR) enables magnified images with a high space resolution that might be used as a non-invasive and non-destructive technique for the quantitative analysis of medical images, in particular the histomorphometry (HMM) of bony mass. In the preprocessing of such images, conventional operations such as binarization and morphological filtering are used before calculating the stereological parameters related, for example, to the trabecular bone microarchitecture. However, there is no standardization of methods for HMM based on μCT images, especially the ones obtained with SR X-ray. Notwithstanding the several uses of artificial neural networks (ANNs) in medical imaging, their application to the HMM of SR-μCT medical images is still incipient, despite the potential of both techniques. The contribution of this paper is the assessment and comparison of well-known training algorithms as well as the proposal of training strategies (combinations of training algorithms, sub-image kernel and symmetry information) for feed-forward ANNs in the task of bone pixels recognition in SR-μCT medical images. For a quantitative comparison, the results of a cross validation and a statistical analysis of the results for 36 training strategies are presented. The ANNs demonstrated both very low mean square errors in the validation, and good quality segmentation of the image of interest for application to HMM in SR-μCT medical images.

Design and development of teaching system for medical image network

Design and development of teaching system for medical image network

Improved beam uniformity in multimode fibers using piezoelectric-based spatial mode scrambling for medical applications

We improve the beam uniformity passing through multimode fibers using piezoelectric-based spatial mode scrambling. Both fiber squeezing and stretching techniques are applied and compared. A more than sixfold difference in the power intensity crossing the output beam profile without mode scrambling can be reduced to a <10% variation. The efficiency of collecting useful optical power for detection is also significantly improved. Such modules can find various applications in medical imaging, disease diagnosis, and local area networks.