Age-related macular degeneration (AMD) is a common threat to vision. While classification of disease stages is critical to understanding disease risk and progression, several systems based on color ...fundus photographs are known. Most of these require in-depth and time-consuming analysis of fundus images. Herein, we present an automated computer-based classification algorithm.
Algorithm development for AMD classification based on a large collection of color fundus images. Validation is performed on a cross-sectional, population-based study.
We included 120 656 manually graded color fundus images from 3654 Age-Related Eye Disease Study (AREDS) participants. AREDS participants were >55 years of age, and non-AMD sight-threatening diseases were excluded at recruitment. In addition, performance of our algorithm was evaluated in 5555 fundus images from the population-based Kooperative Gesundheitsforschung in der Region Augsburg (KORA; Cooperative Health Research in the Region of Augsburg) study.
We defined 13 classes (9 AREDS steps, 3 late AMD stages, and 1 for ungradable images) and trained several convolution deep learning architectures. An ensemble of network architectures improved prediction accuracy. An independent dataset was used to evaluate the performance of our algorithm in a population-based study.
κ Statistics and accuracy to evaluate the concordance between predicted and expert human grader classification.
A network ensemble of 6 different neural net architectures predicted the 13 classes in the AREDS test set with a quadratic weighted κ of 92% (95% confidence interval, 89%–92%) and an overall accuracy of 63.3%. In the independent KORA dataset, images wrongly classified as AMD were mainly the result of a macular reflex observed in young individuals. By restricting the KORA analysis to individuals >55 years of age and prior exclusion of other retinopathies, the weighted and unweighted κ increased to 50% and 63%, respectively. Importantly, the algorithm detected 84.2% of all fundus images with definite signs of early or late AMD. Overall, 94.3% of healthy fundus images were classified correctly.
Our deep learning algoritm revealed a weighted κ outperforming human graders in the AREDS study and is suitable to classify AMD fundus images in other datasets using individuals >55 years of age.
In the framework of nondestructive testing and evaluation, Lorentz force evaluation (LFE) is a method for reconstructing defects in electrically conducting laminated composites. In this paper, we ...propose a new inverse calculation strategy for LFE based on a stochastic optimization, the differential evolution (DE) algorithm. We determined the optimal control parameters for the DE and assessed its performance based on simulated and measured data. The results show that the depth of the defect was estimated correctly for all of the data sets that we evaluated. The geometry was reconstructed with errors of less than 4% relative to the size of the defect. The proposed scheme was robust against noise and distortions in the data measurements. We conclude that the proposed reconstruction scheme is a promising method for solving the inverse problem in LFE.
Permanent Magnet Modeling for Lorentz Force Evaluation Mengelkamp, Judith; Ziolkowski, Marek; Weise, Konstantin ...
IEEE transactions on magnetics,
2015-July, 2015-7-00, 20150701, Volume:
51, Issue:
7
Journal Article
Lorentz force evaluation (LFE) is a technique to reconstruct defects in electrically conductive materials. The accuracy of the forward and inverse solution highly depends on the applied model of the ...permanent magnet. The resolution of the technique relies upon the shape and size of the permanent magnet. Furthermore, the application of an existing forward solution requires an analytic integral of the magnetic flux density. Motivated by these aspects, we propose a magnetic dipoles model (MDM), in which the permanent magnet is substituted with an assembly of magnetic dipoles. This approach allows modeling of magnets of arbitrary shape by appropriate positioning of the dipoles, and the integral can be expressed by elementary mathematical functions. We apply the MDM to cuboidal-shaped and cylindrical-shaped magnets and evaluate the obtained magnetic flux density by comparing it to reference solutions. We consider distances of 2-6 mm to the permanent magnet. The representation of a cuboidal magnet with 832 dipoles yields a maximum error of 0.02% between the computed magnetic field of the MDM and the reference solution. Comparable accuracy for the cylindrical magnet is achieved with 1890 dipoles. In addition, we embed the MDM of the cuboidal magnet into an existing forward solution for LFE and find that the errors of the magnetic flux density are partly compensated by the forward calculations. We conclude that our modeling approach can be used to determine the most efficient MDMs for LFE.
The detection and reconstruction of fatigue fractures is of great interest in quality assurance. In the framework of nondestructive testing, Lorentz force evaluation (LFE) is an evaluation technique ...to estimate flaws in electrically conductive materials based on measured Lorentz forces. In the forward solution for LFE, a defect can be interpreted as a distributed current source. This has motivated the authors to propose current density reconstructions (CDRs) calculated with minimum norm estimates to estimate defect geometries. The L
1
and L
2
norms tend to produce a solution which is either very focused or very smeared. To balance these constraints, the general L
p
norm with 1 ≤ p ≤ 2 was used and the inverse solutions compared. This approach was applied to measured data obtained from a laminated composite and simulated data from a monolithic material. The results show that the L
1.5
norm provides the most accurate inverse solutions. The location and extent of the defect are determined with an error of 15 % relative to the size of the defect. The depth estimation has a deviation of 50 %. It can be concluded that CDRs are a powerful method to reconstruct and characterize defects in LFE.
Full text
Available for:
BFBNIB, GIS, IJS, KISLJ, NUK, PNG, UL, UM, UPUK
Purpose - The purpose of this paper is to present a novel electromagnetic non-destructive evaluation technique, so called Lorentz force eddy current testing (LET). This method can be applied for the ...detection and reconstruction of defects lying deep inside a non-magnetic conducting material. Design/methodology/approach - In this paper the technique is described in general as well as its experimental realization. Besides that, numerical simulations are performed and compared to experimental data. Using the output data of measurements and simulations, an inverse calculation is performed in order to reconstruct the geometry of a defect by means of sophisticated optimization algorithms. Findings - The results show that measurement data and numerical simulations are in a good agreement. The applied inverse calculation methods allow to reconstruct the dimensions of the defect in a suitable accuracy. Originality/value - LET overcomes the frequency dependent skin-depth of traditional eddy current testing due to the use of permanent magnets and low to moderate magnetic Reynolds numbers (0.1-1). This facilitates the possibility to detect subsurface defects in conductive materials.
To identify the scalp projections of the underlying sources of neural activity based on recorded electroencephalographic (EEG) signals, the multi-dimensional decomposition models Parallel Factor ...Analysis (PARAFAC) and Parallel Factor Analysis 2 (PARAFAC2) have recently attained interest. We evaluate the models based on synthetic EEG data, because this allows an objective assessment by comparing the estimated projections to the parameters of the sources. We simulate EEG data using the EEG forward solution and focus on dynamic sources that change their spatial projection over time. Recently, this type of signal has been identified as the dominant type of signal, e. g. in measurements of visual evoked potentials. Further, we develop a method to objectively evaluate the decomposition models. We show that the decomposition models reconstruct the scalp projections successfully from data with low signal-to-noise ratio (SNR). They perform best if the number of calculated components (model order) equals the number of sources.