High Performance Medical Image Research Articles

High-Resolution X-ray Image from Copper-Based Perovskite Hybrid Polymer.

Cs3Cu2I5 nanocrystals (NCs) are considered to be promising materials due to their high photoluminescence efficiency, lack of lead toxicity, and X-ray responsiveness. However, during the crystallization process, NCs are prone to agglomeration and exhibit uneven size distribution, resulting in several light scattering that severely affect their imaging resolution. Herein, we successfully developed a high-resolution scintillator film by growing copper-based perovskite NCs within a hybrid polymer matrix. By leveraging the ingenious integration of polyvinylidene fluoride (PVDF) and polymethyl methacrylate (PMMA), the size and distribution uniformity of Cs3Cu2I5 NCs can be effectively controlled. Consequently, a high spatial resolution of 14.3 lp mm-1 and a low detection limit of 105 nGy s-1 are achieved, and the scintillator film has excellent flexibility and stability. These results highlight the promising application of Cs3Cu2I5 scintillator films in low-cost, flexible, and high-performance medical imaging.

A review of research and development of semi-supervised learning strategies for medical image processing

Accurate and robust segmentation of organs or lesions from medical images plays a vital role in many clinical applications such as diagnosis and treatment planning. With the massive increase in labeled data, deep learning has achieved great success in image segmentation. However, for medical images, the acquisition of labeled data is usually expensive because generating accurate annotations requires expertise and time, especially in 3D images. To reduce the cost of labeling, many approaches have been proposed in recent years to develop a high-performance medical image segmentation model to reduce the labeling data. For example, combining user interaction with deep neural networks to interactively perform image segmentation can reduce the labeling effort. Self-supervised learning methods utilize unlabeled data to train the model in a supervised manner, learn the basics and then perform knowledge transfer. Semi-supervised learning frameworks learn directly from a limited amount of labeled data and a large amount of unlabeled data to get high quality segmentation results. Weakly supervised learning approaches learn image segmentation from borders, graffiti, or image-level labels instead of using pixel-level labeling, which reduces the burden of labeling. However, the performance of weakly supervised learning and self-supervised learning is still limited on medical image segmentation tasks, especially on 3D medical images. In addition to this, a small amount of labeled data and a large amount of unlabeled data are more in line with actual clinical scenarios. Therefore, semi-supervised learning strategies become very important in the field of medical image processing.

Lattice-Gradient Perovskite KTaO3 Films for an Ultrastable and Low-Dose X-Ray Detector.

Conventional indirect X-ray detectors employ scintillating phosphors to convert X-ray photons into photodiode-detectable visible photons, leading to low conversion efficiencies, low spatial resolutions, and optical crosstalk. Consequently, X-ray detectors that directly convert photons into electric signals have long been desired for high-performance medical imaging and industrial inspection. Although emerging hybrid inorganic-organic halide perovskites, such as CH3 NH3 PbI3 and CH3 NH3 PbBr3 , exhibit high sensitivity, they have salient drawbacks including structural instability, ion motion, and the use of toxic Pb. Here, this work reports an ultrastable, low-dose X-ray detector comprising KTaO3 perovskite films epitaxially grown on a Nb-doped strontium titanate substrate using a low-cost solution method. The detector exhibits a stable photocurrent under high-dose irradiation, high-temperature (200 °C), and aqueous conditions. Moreover, the prototype KTaO3 -film-based detector exhibits a 150-fold higher sensitivity (3150 µC Gyair -1 cm-2 ) and 150-fold lower detection limit (<40 nGyair s-1 ) than those of commercial α-Se-based direct detectors. Systematic investigations reveal that the high stability of the detector originates from the strong covalent bonds within the KTaO3 film, whereas the low detection limit is due to a lattice-gradient-driven built-in electric field and the high insulating property of KTaO3 film. This study unveils a new path toward the fabrication of green, stable, and low-dose X-ray detectors using oxide perovskite films, which have significant application potential in medical imaging and security operations.

Data augmentation based on multiple oversampling fusion for medical image segmentation

A high-performance medical image segmentation model based on deep learning depends on the availability of large amounts of annotated training data. However, it is not trivial to obtain sufficient annotated medical images. Generally, the small size of most tissue lesions, e.g., pulmonary nodules and liver tumours, could worsen the class imbalance problem in medical image segmentation. In this study, we propose a multidimensional data augmentation method combining affine transform and random oversampling. The training data is first expanded by affine transformation combined with random oversampling to improve the prior data distribution of small objects and the diversity of samples. Secondly, class weight balancing is used to avoid having biased networks since the number of background pixels is much higher than the lesion pixels. The class imbalance problem is solved by utilizing weighted cross-entropy loss function during the training of the CNN model. The LUNA16 and LiTS17 datasets were introduced to evaluate the performance of our works, where four deep neural network models, Mask-RCNN, U-Net, SegNet and DeepLabv3+, were adopted for small tissue lesion segmentation in CT images. In addition, the small tissue segmentation performance of the four different deep learning architectures on both datasets could be greatly improved by incorporating the data augmentation strategy. The best pixelwise segmentation performance for both pulmonary nodules and liver tumours was obtained by the Mask-RCNN model, with DSC values of 0.829 and 0.879, respectively, which were similar to those of state-of-the-art methods.

Cutting-Edge Imaging of Cardiac Metastases from Neuroendocrine Tumors: Lesson from a Case Series

With the increasing availability of high-performance medical imaging for the management of patients with neuroendocrine tumors (NETs), a progressive growth of asymptomatic and incidentally detected cardiac metastases (CMs) has been observed in the recent years. In clinical practice, CMs of NENs are often incidentally detected by whole-body 68Ga-labeled somatostatin analogs or 18F-fluorodihydroxyphenylalanine positron emission tomography/computed tomography, and afterwards accurately characterized by cardiac magnetic resonance (CMR) and/or gated cardiac computed tomography when CMR is contraindicated or not available. The interpreting physician should familiarize with the main imaging features of CM, a finding that may be encountered in NETs patients more than previously thought. Herein, we present a case series of four patients with CMs from small-intestine NETs highlighting strengths and weaknesses of a multimodality imaging approach in clinical practice.

High-performance medical image secret sharing using super-resolution for CAD systems

High-performance medical image secret sharing using super-resolution for CAD systems

In Situ Fabrication of Cs3Cu2I5: Tl Nanocrystal Films for High-Resolution and Ultrastable X-ray Imaging.

Cs3Cu2I5 nanocrystals (NCs) are considered to be promising materials due to their high photoluminescence efficiency and X-ray hardness. However, the present strategy depends on tedious fabrication with excessive chemical waste. The evasive iodide ion dissociation, inadaptable ligand system, low stability, and relatively low light yield severely impede their applications. Herein, we develop an in situ fabrication strategy for a flexible and large-area Tl-doped Cs3Cu2I5 NC-polymer composite scintillation film with a high light yield (∼48800 photons/MeV) and improved stability. Tween 80 and phosphinic acid successfully inhibit the oxidation of iodide ions, and the films can be stored for at least six months. As a result, a high spatial resolution of 16.3 lp mm-1 and a low detection limit of 305 nGyair s-1 were achieved. A radioluminescence intensity of >80% was maintained after a total irradiation dose of 604.8 Gy. These results indicate the promising application of these copper halide NCs in low-cost, flexible, and high-performance medical imaging.

An Analysis on Ensemble Learning Optimized Medical Image Classification With Deep Convolutional Neural Networks

Novel and high-performance medical image classification pipelines are heavily utilizing ensemble learning strategies. The idea of ensemble learning is to assemble diverse models or multiple predictions and, thus, boost prediction performance. However, it is still an open question to what extent as well as which ensemble learning strategies are beneficial in deep learning based medical image classification pipelines. In this work, we proposed a reproducible medical image classification pipeline for analyzing the performance impact of the following ensemble learning techniques: Augmenting, Stacking, and Bagging. The pipeline consists of state-of-the-art preprocessing and image augmentation methods as well as 9 deep convolution neural network architectures. It was applied on four popular medical imaging datasets with varying complexity. Furthermore, 12 pooling functions for combining multiple predictions were analyzed, ranging from simple statistical functions like unweighted averaging up to more complex learning-based functions like support vector machines. Our results revealed that Stacking achieved the largest performance gain of up to 13% F1-score increase. Augmenting showed consistent improvement capabilities by up to 4% and is also applicable to single model based pipelines. Cross-validation based Bagging demonstrated significant performance gain close to Stacking, which resulted in an F1-score increase up to +11%. Furthermore, we demonstrated that simple statistical pooling functions are equal or often even better than more complex pooling functions. We concluded that the integration of ensemble learning techniques is a powerful method for any medical image classification pipeline to improve robustness and boost performance.

An Analysis on Ensemble Learning optimized Medical Image Classification with Deep Convolutional Neural Networks

Novel and high-performance medical image classification pipelines are heavily utilizing ensemble learning strategies. The idea of ensemble learning is to assemble diverse models or multiple predictions and, thus, boost prediction performance. However, it is still an open question to what extent as well as which ensemble learning strategies are beneficial in deep learning based medical image classification pipelines. In this work, we proposed a reproducible medical image classification pipeline for analyzing the performance impact of the following ensemble learning techniques: Augmenting, Stacking, and Bagging. The pipeline consists of state-of-the-art preprocessing and image augmentation methods as well as 9 deep convolution neural network architectures. It was applied on four popular medical imaging datasets with varying complexity. Furthermore, 12 pooling functions for combining multiple predictions were analyzed, ranging from simple statistical functions like unweighted averaging up to more complex learning-based functions like support vector machines. Our results revealed that Stacking achieved the largest performance gain of up to 13% F1-score increase. Augmenting showed consistent improvement capabilities by up to 4% and is also applicable to single model based pipelines. Cross-validation based Bagging demonstrated significant performance gain close to Stacking, which resulted in an F1-score increase up to +11%. Furthermore, we demonstrated that simple statistical pooling functions are equal or often even better than more complex pooling functions. We concluded that the integration of ensemble learning techniques is a powerful method for any medical image classification pipeline to improve robustness and boost performance.

Phase structures and electrical properties of Sm doped PSN-PMN-PT ceramics

A series of 2 mol% Sm doped 0.15Pb(Sc1/2Nb1/2)O3-(0.85-x)Pb(Mg1/3Nb2/3) O3-xPbTiO3 (Sm-PSN-PMN-xPT, x = 0.32, 0.33, 0.35 and 0.37) piezoelectric ceramics were fabricated by solid-state method, and the detailed crystal structure, electrical properties and thermal stability were investigated. The Rietveld fitting XRD results show the ceramics with x = 0.32 and 0.33 possess concomitant monoclinic (Mc) and rhombohedral (R) phases at room temperature, while coexistence phases of tetragonal (T) and monoclinic (Mc) are observed in x = 0.35 and 0.37 ceramics. The high piezoelectric coefficient d33 of 815 pC/N and electromechanical coupling coefficient kp of 0.61, and a large unipolar strain Smax of 0.134% accompanied by a small strain hysteresis of 4.74% are obtained in x = 0.33 ceramics. Through the Rayleigh analysis, these high properties are mainly derived from intrinsic contributions (lattice distortion and reversible domain walls displacement). It is particularly important that in the temperature range from 30 °C to 110 °C, the variations of d33 and kp are only 8% and 13%, respectively, demonstrating insensitive temperature-dependent behavior of x = 0.33 ceramics, which can be further confirmed by the temperature dependence of εr × Pr, showing a variation of less than 14%. Our results indicated the Sm doped PSN-PMN-PT ceramics have potential applications in high performance medical ultrasonic imaging transducers and actuators.

End-to-end privacy preserving deep learning on multi-institutional medical imaging

Using large, multi-national datasets for high-performance medical imaging AI systems requires innovation in privacy-preserving machine learning so models can train on sensitive data without requiring data transfer. Here we present PriMIA (Privacy-preserving Medical Image Analysis), a free, open-source software framework for differentially private, securely aggregated federated learning and encrypted inference on medical imaging data. We test PriMIA using a real-life case study in which an expert-level deep convolutional neural network classifies paediatric chest X-rays; the resulting model’s classification performance is on par with locally, non-securely trained models. We theoretically and empirically evaluate our framework’s performance and privacy guarantees, and demonstrate that the protections provided prevent the reconstruction of usable data by a gradient-based model inversion attack. Finally, we successfully employ the trained model in an end-to-end encrypted remote inference scenario using secure multi-party computation to prevent the disclosure of the data and the model. Gaining access to medical data to train AI applications can present problems due to patient privacy or proprietary interests. A way forward can be privacy-preserving federated learning schemes. Kaissis, Ziller and colleagues demonstrate here their open source framework for privacy-preserving medical image analysis in a remote inference scenario.

Machine Learning for Computer-aided Diagnosis.

Machine learning algorithms are to analyze any dataset to extract data-driven model, prediction rule, or decision rule from the dataset. Various machine learning algorithms are now used to develop high-performance medical image processing systems such as computer-aided detection (CADe) system which detects clinically significant objects from medical images and computer-aided diagnosis (CADx) system which quantifies malignancy of manually or automatically detected clinical objects. In this paper, we introduce some applications of machine learning algorithms to the development of medical image processing system.

Resource estimation in high performance medical image computing.

Medical imaging analysis processes often involve the concatenation of many steps (e.g., multi-stage scripts) to integrate and realize advancements from image acquisition, image processing, and computational analysis. With the dramatic increase in data size for medical imaging studies (e.g., improved resolution, higher throughput acquisition, shared databases), interesting study designs are becoming intractable or impractical on individual workstations and servers. Modern pipeline environments provide control structures to distribute computational load in high performance computing (HPC) environments. However, high performance computing environments are often shared resources, and scheduling computation across these resources necessitates higher level modeling of resource utilization. Submission of 'jobs' requires an estimate of the CPU runtime and memory usage. The resource requirements for medical image processing algorithms are difficult to predict since the requirements can vary greatly between different machines, different execution instances, and different data inputs. Poor resource estimates can lead to wasted resources in high performance environments due to incomplete executions and extended queue wait times. Hence, resource estimation is becoming a major hurdle for medical image processing algorithms to efficiently leverage high performance computing environments. Herein, we present our implementation of a resource estimation system to overcome these difficulties and ultimately provide users with the ability to more efficiently utilize high performance computing resources.

A Study of NetCDF as an Approach for High Performance Medical Image Storage

The spread of telemedicine systems increases every day. The systems and PACS based on DICOM images has become common. This rise reflects the need to develop new storage systems, more efficient and with lower computational costs. With this in mind, this article discusses a study for application in NetCDF data format as the basic platform for storage of DICOM images. The study case comparison adopts an ordinary database, the HDF5 and the NetCDF to storage the medical images. Empirical results, using a real set of images, indicate that the time to retrieve images from the NetCDF for large scale images has a higher latency compared to the other two methods. In addition, the latency is proportional to the file size, which represents a drawback to a telemedicine system that is characterized by a large amount of large image files.

High-Performance Medical Imaging Informatics

High-Performance Medical Imaging Informatics

Towards Building High Performance Medical Image Management System for Clinical Trials.

Medical image based biomarkers are being established for therapeutic cancer clinical trials, where image assessment is among the essential tasks. Large scale image assessment is often performed by a large group of experts by retrieving images from a centralized image repository to workstations to markup and annotate images. In such environment, it is critical to provide a high performance image management system that supports efficient concurrent image retrievals in a distributed environment. There are several major challenges: high throughput of large scale image data over the Internet from the server for multiple concurrent client users, efficient communication protocols for transporting data, and effective management of versioning of data for audit trails. We study the major bottlenecks for such a system, propose and evaluate a solution by using a hybrid image storage with solid state drives and hard disk drives, RESTful Web Services based protocols for exchanging image data, and a database based versioning scheme for efficient archive of image revision history. Our experiments show promising results of our methods, and our work provides a guideline for building enterprise level high performance medical image management systems.

Signal and noise transfer properties of CMOS based active pixel flat panel imager coupled to structured CsI:Tl

Complementary metal-oxide-semiconductors (CMOS) active pixel sensors can be optically coupled to CsI:Tl phosphors forming a indirect active pixel flat panel imager (APFPI) for high performance medical imaging. The aim of this work is to determine the x-ray imaging capabilities of CMOS-based APFPI and study the signal and noise transfer properties of CsI:Tl phosphors. Three different CsI:Tl phosphors from two different vendors have been used to produce three system configurations. The performance of each system configuration has been studied in terms of the modulation transfer function (MTF), noise power spectra, and detective quantum efficiency (DQE) in the mammographic energy range. A simple method to determine quantum limited systems in this energy range is also presented. In addition, with aid of monochromatic synchrotron radiation, the effect of iodine characteristic x-rays of the CsI:Tl on the MTF has been determined. A Monte Carlo simulation of the signal transfer properties of the imager is also presented in order to study the stages that degrade the spatial resolution of our current system. The effect of using substrate patterning during the growth of CsI:Tl columnar structure was also studied, along with the effect of CsI:Tl fixed pattern noise due to local variations in the scintillation light. CsI:Tl fixed pattern noise appears to limit the performance of our current system configurations. All the system configurations are quantum limited at 0.23 microC/kg with two of them having DQE (0) equal to 0.57. Active pixel flat panel imagers are shown to be digital x-ray imagers with almost constant DQE throughout a significant part of their dynamic range and in particular at very low exposures.

High-Performance Medical Image Registration Using New Optimization Techniques

Optimization of a similarity metric is an essential component in intensity-based medical image registration. The increasing availability of parallel computers makes parallelizing some registration tasks an attractive option to increase speed. In this paper, two new deterministic, derivative-free, and intrinsically parallel optimization methods are adapted for image registration. DIviding RECTangles (DIRECT) is a global technique for linearly bounded problems, and multidirectional search (MDS) is a recent local method. The performance of DIRECT, MDS, and hybrid methods using a parallel implementation of Powell's method for local refinement, are compared. Experimental results demonstrate that DIRECT and MDS are robust, accurate, and substantially reduce computation time in parallel implementations.

Image quality degradation by light scattering processes in high performance display devices for medical imaging

This thesis addresses the characterization of light scattering processes that degrade image quality in high performance electronic display devices for radiography. Using novel experimental and computational tools, we study the lateral diffusion of light in emissive displays that causes extensive veiling glare and significant reduction of the contrast. In addition, we examine the deleterious effects of ambient light reflections that affect the contrast of low luminance regions, and superimpose unwanted structured signal. To model veiling glare and display reflections, we use a Monte Carlo light transport simulation code, DETECT‐II, that tracks individual photons through multiple scattering events. The simulation accounts for the photon polarization state, and provides descriptions for rough surfaces and thin film coatings. A new experimental method to measure veiling glare is described based on a conic collimated probe that minimizes unwanted contributions from bright areas. We show that veiling glare ratios in the order of a few hundreds can be measured with an uncertainty of a few percent. For a high performance medical imaging monitor with anti‐reflective coating, the veiling glare ratio for a 1 cm diameter dark spot was measured to be 240. Finally, we introduce experimental techniques for measurements of display reflectance, and we compare measured reflection coefficients with Monte Carlo estimates. In spite of having comparable reflection coefficients, the low maximum luminance of current cathode‐ray tube devices worsens the effect of ambient light reflections when compared to radiographic film. Flat panel technologies can perform even better than film due to a thin faceplate, increased light absorption, and high brightness.

High performance medical image processing in client/server-environments

As 3D scanning devices like computer tomography (CT) or magnetic resonance imaging (MRI) become more widespread, there is also an increasing need for powerful computers that can handle the enormous amounts of data with acceptable response times. We describe an approach to parallelize some of the more frequently used image processing operators on distributed memory architectures. It is desirable to make such specialized machines accessible on a network, in order to save costs by sharing resources. We present a client/server approach that is specifically tailored to the interactive work with volume data. Our image processing server implements a volume visualization method that allows the user to assess the segmentation of anatomical structures. We can enhance the presentation by combining the volume visualizations on a viewing station with additional graphical elements, which can be manipulated in real-time. The methods presented were verified on two applications for different domains.