Search

Background Currently there is an ever increasing interest in Lu-177 targeted radionuclide therapies, which target neuro-endocrine and prostate tumours. For a patient-specific treatment, an individual dosimetry based on SPECT/CT imaging is necessary. The aim of this study is to introduce a dosimetry method, where dose voxel kernels (DVK) are predicted by a neural network. Methods Kidneys are considered one of the most critical organs in any radionuclide therapy. Hence we chose kidneys of 26 patients, who underwent Lu-177-DOTATOC or PSMA therapy, as target organs for our dosimetric method. First of all, density kernels with a size of 9 × 9 × 9 voxels were considered, and the corresponding DVKs were calculated by Monte Carlo simulations. These kernels were used to train a neural network (NN), which received a density kernel as input and predicted a DVK at the output. To predict the dose distribution of an entire kidney, the organ had to be partitioned into a large number of density kernels. Afterwards the DVKs were predicted by a trained NN, and employed to reconstruct the whole-organ dose distribution by convolution with the activity distribution. This method was compared to the standard method where the activity distribution is convolved with a DVK based on a homogeneous soft tissue kernel. Results The number of training kernels amounted to 52,274 density kernels with corresponding MC-derived DVKs. The results serve as proof of principle of the newly proposed method and showed that the NN approach yielded superior results compared to the standard method with no additional computational effort. Conclusion The NN approach is an accurate and highly competitive dosimetric method to precisely estimate absorbed radiation dose in critical organs like kidneys in clinical routine. To further improve the results, a larger number of DVKs needs to be computed by Monte Carlo simulations. An extension of the method to other organs is easily conceivable.

In this study, an automated machine learning approach for the segmentation of MS lesions from multi- modal magnetic resonance images (mmMRI) is presented. The method is based on a U-Net like convolutional neural network (CNN) for 2D slice-based segmentation of 3D brain MRI volumes. The different modalities are encoded in sep- arate downsampling channels. The skip connections input feature maps to multi-scale feature fusion blocks at every stage of the network. These are followed by multi-scale feature upsampling blocks, which use the information from lesion shape and location. The CNN is evaluated on two publicly available datasets: The ISBI 2015 longitudinal MS lesion segmentation challenge dataset which contains 14 MS patients and the MICCAI 2016 MSSEG challenge dataset consisting of 15 MS patients. Regarding the ISBI Challenge, the proposed method was among the top performing ap- proaches to which open-access papers are available. The MICCAI dataset served to evaluate the robustness of the architecture against scanner variability and to show the improvements in performance after transfer learning.

Patients with advanced prostate cancer are suitable candidates for [ 13 patients with histologically confirmed metastasized castration-resistant prostate cancer were treated with a total of 18 cycles of [ The average dose values yielded the following results: 2.59 ± 0.63 Gy (1.67-3.92 Gy) for the kidneys, 0.79 ± 0.46 Gy (0.31-1.90 Gy) for the spleen and 11.00 ± 11.97 Gy (1.28-49.10 Gy) for all tracer-positive tumor lesions. A trend towards higher TTD was observed in patients with Gleason Scores 8 compared to Gleason Scores ≤ 8 and in lymph node metastases compared to bone metastases. A significant correlation was determined between the serum-PSA level before RLT and the TTD (r = -0.57, p 0.05), as well as between the TTD with the percentage change of serum-PSA levels before and after therapy was observed (r = -0.57, p 0.05). Patients with higher total tumor volumes of PSMA-positive lesions demonstrated significantly lower kidney average dose values (r = -0.58, p 0.05). The presented novel Monte Carlo based voxel-wise dosimetry calculates a patient specific whole-body dose distribution, thus taking into account individual anatomies and tissue compositions showing promising results for the estimation of radiation doses of normal organs and PSMA-positive tumor lesions.ZIEL: Patienten mit fortgeschrittenem Prostatakrebs sind geeignete Kandidaten für die [ 13 Patienten mit histologisch gesichertem metastasiertem kastrationsresistentem Prostatakrebs wurden mit insgesamt 18 Zyklen [ Die mittleren Dosiswerte wurden wie folgt bestimmt: 2,59 ± 0,63 Gy (1,67–3,92 Gy) für die Nieren, 0,79 ± 0,46 Gy (0,31–1,90 Gy) für die Milz und 11,00 ± 11,97 Gy (1,28–49,10 Gy) für alle Tracer-positiven Tumorläsionen. Ein Trend zu höheren totalen Tumordosen wurde in Patienten mit Gleason-Scores 8 im Vergleich zu Gleason-Scores ≤ 8 und in Patienten mit Lymphknotenmetastasen im Vergleich zu Knochenmetastasen beobachtet. Eine signifikante Korrelation wurde zwischen Serum-PSA-Werten vor Radioligandentherapie und der totalen Tumordosis ermittelt (r = –0,57; p 0,05) sowie zwischen der totalen Tumordosis und der prozentualen Änderung der Serum-PSA-Werte vor und nach Therapie beobachtet (r = –0,57; p 0,05). Patienten mit höherem totalem Tumorvolumen der PSMA-positiven Läsionen zeigten signifikant geringere mittlere Nierendosen (r = –0,58; p 0,05). Die vorgestellte neuartige Monte-Carlo-basierte voxelweise Dosimetrie berechnet eine patientenspezifische Ganzkörper-Dosisverteilung, die individuelle Gewebezusammensetzungen berücksichtigt und vielversprechende Resultate für die Bestimmung der absorbierten Dosis von Organen und PSMA-positiven Tumorläsionen zeigt.

Background Currently, there is a high interest in 177 Lu targeted radionuclide therapies, which could be attributed to favorable results obtained from 177 Lu compounds targeting neuro-endocrine and prostate tumors. SPECT based dosimetry could be used for deriving dose values for individual voxels, as is the standard in external-beam radiation-therapy (EBRT). For this a time-activity-curve (TAC) at voxel resolution and also a voxel-wise modeling of radiation energy deposition are necessary. But a voxel-wise determination of TACs is problematic, since several confounding factors exist, such as e.g. poor count-statistics or registration inaccuracies, which add noise to the observed activity states. A particle filter (PF) is a class of methods which applies regularization based on a model of the temporal evolution of activity states. The aim of this study is to introduce the application of PFs for de-noising of per-voxel time-activity curves. Methods We applied a PF for de-noising the TACs of 26 patients, who underwent 177 Lu-DOTATOC or -PSMA therapy. The TACs were obtained from fully-quantitative, serial SPECT(/CT) data, acquired at 4 h, 24 h, 48 h, 72 h p.i. The model used in the PF was a mono-exponential decay and its free parameters were determined based on objective criteria. The time-integrated activities (TIA) resulting from the PF (PFF) were compared to the results of a mono-exponential fit (SF) of individual voxels in several volumes of interest (kidneys, spleen, tumors). Additionally, an organ-averaged TIA was derived from whole-organ VOIs and subsequent curve-fitting. This whole-organ TIA was also compared to the whole-organ TIAs obtained from summation of the voxel-wise TIAs from PFF and SF. Results The number of particles was set to 1000. Optimal values for noise of observations and noise of the model were 0.25 and 0.5, respectively. The deviation of whole-organ TIAs from conventional organ-based dosimetry and the summation of the voxel-wise TIAs was substantial for SF (kidneys − 22.3%, spleen − 49.6%, tumor − 60.0%), as well as for PFF (kidneys − 37.1%, spleen − 57.9%, tumor − 70.9%). The distribution of voxel-wise half-lives resulting from the PFF method was considerably closer to the organ-averaged value, and the number of implausibly long half-lives ( > physical HL) was reduced. Conclusion The PFF leads to voxel-wise half-lives, which are more plausible than those resulting from SF. However, one has to admit that voxel-wise fitting generally leads to considerable deviations from the organ-averaged TIA as obtained by conventional whole-organ evaluation. Unfortunately, we did not have ground-truth TIA of our patient data and proper ground-truth could even be impossible to obtain. Nevertheless, there are strong indicators that particle filtering can be used for reducing voxel-wise TAC noise.

Background Patients with advanced neuroendocrine tumors (NETs) of the midgut are suitable candidates for 177Lu-DOTATOC therapy. Integrated SPECT/CT systems have the potential to help improve the accuracy of patient-specific tumor dosimetry. Dose estimations to target organs are generally performed using the Medical Internal Radiation Dose scheme. We present a novel Monte Carlo-based voxel-wise dosimetry approach to determine organ- and tumor-specific total tumor doses (TTD). Methods A cohort of 14 patients with histologically confirmed metastasized NETs of the midgut (11 men, 3 women, 62.3 ± 11.0 years of age) underwent a total of 39 cycles of 177Lu-DOTATOC therapy (mean 2.8 cycles, SD ± 1 cycle). After the first cycle of therapy, regions of interest were defined manually on the SPECT/CT images for the kidneys, the spleen, and all 198 tracer-positive tumor lesions in the field of view. Four SPECT images, taken at 4 h, 24 h, 48 h and 72 h after injection of the radiopharmaceutical, were used to determine their effective half-lives in the structures of interest. The absorbed doses were calculated by a three-dimensional dosimetry method based on Monte Carlo simulations. TTD was calculated as the sum of all products of single tumor doses with single tumor volumes divided by the sum of all tumor volumes. Results The average dose values per cycle were 3.41 ± 1.28 Gy (1.91–6.22 Gy) for the kidneys, 4.40 ± 2.90 Gy (1.14–11.22 Gy) for the spleen, and 9.70 ± 8.96 Gy (1.47–39.49 Gy) for all 177Lu-DOTATOC-positive tumor lesions. Low- and intermediate-grade tumors (G 1–2) absorbed a higher TTD compared to high-grade tumors (G 3) (signed-rank test, p = p = 0.01). Higher TTD resulted in a significant decrease of CgA after therapy. Conclusion These results suggest that Monte Carlo-based voxel-wise dosimetry is a very promising tool for predicting the absorbed TTD based on histological and clinical parameters.

Human learning and intelligence work differently from the supervised pattern recognition approach adopted in most deep learning architectures. Humans seem to learn rich representations by exploration and imitation, build causal models of the world, and use both to flexibly solve new tasks. We suggest a simple but effective unsupervised model which develops such characteristics. The agent learns to represent the dynamical physical properties of its environment by intrinsically motivated exploration, and performs inference on this representation to reach goals. For this, a set of self-organizing maps which represent state-action pairs is combined with a causal model for sequence prediction. The proposed system is evaluated in the cartpole environment. After an initial phase of playful exploration, the agent can execute kinematic simulations of the environment's future, and use those for action planning. We demonstrate its performance on a set of several related, but different one-shot imitation tasks, which the agent flexibly solves in an active inference style., Comment: 27 pages, 5 figures

The paper introduces a novel approach for the extraction of physically meaningful thermal component time series during the manufacturing of casting parts. We treat their extraction as Blind Source Separation (BSS) problem by exploiting process-related prior knowledge. Our proposed method arranges temperature time series into a data matrix, which is then decomposed by Non-negative Matrix Factorization (NMF). The latter is guided by a knowledge-based strategy, which initializes the NMF component matrix with time curves designed according to basic physical processes. The temperature time series encompass exclusively non-negative data. Hence NMF lends itself a natural choice as it does not impose mathematical constraints that lack any immediate physical interpretation. We show how to extract components linked to physical phenomena that typically occur during production and cannot be monitored directly. We apply our method to real world data, collected in a foundry during the series production of casting parts for the automobile industry and demonstrate its efficiency.

Face recognition or verification remains a real challenge in the area of pattern recognition and image processing. The image acquisition process is a crucial step in which noise will inevitably be introduced, and in most cases this noise drastically decreases the accuracy of the classification rate of recognition systems, making them ineffective. This paper presents a novel approach to face recognition or verification, which increases the recognition rate in noisy environmental conditions. The latter is achieved by using the intrinsic face mode functions that result from applying a bi-dimensional empirical mode decomposition with Green’s functions in tension to noisy images. Each image is individually decomposed, and noisy modes are discarded or filtered during reconstruction. Then, the extracted modes are used for classification purposes with canonical classifiers such as vector support machines or k-nearest neighbor classifiers. Experimental results show that this method achieves very stable results, almost independently of the amount of noise added to the image, due to the ability of decomposition to capture the noise in the first mode. Classification results using noisy images are at the same level as other algorithms proposed for the same databases but working on clean images and therefore are better than those obtained using classic image filters in noisy images. Moreover, unlike most of the available algorithms, the algorithm proposed in this paper is based on the input data (without the need to adjust parameters), making it transparent to the user. Finally, the proposed new approach achieves good results independently of the type of noise, the level of noise and the type of the database, which is not possible with other classical methods requiring parameter adjustment.

Purpose High dose rate brachytherapy applies intense and destructive radiation. A treatment plan defines radiation source dwell positions to avoid irradiating healthy tissue. The study discusses methods to quantify any positional changes of source locations along the various treatment sessions. Methods Electromagnetic tracking (EMT) localizes the radiation source during the treatment sessions. But in each session the relative position of the patient relative to the filed generator is changed. Hence, the measured dwell point sets need to be registered onto each other to render them comparable. Two point set registration techniques are compared: a probabilistic method called coherent point drift (CPD) and a multidimensional scaling (MDS) technique. Results Both enable using EMT without external registration and achieve very similar results with respect to dwell position determination of the radiation source. Still MDS achieves smaller grand average deviations (CPD-rPSR: MD = 2.55 mm, MDS-PSR: MD = 2.15 mm) between subsequent dwell position determinations, which also show less variance (CPD-rPSR: IQR = 4 mm, MDS-PSR: IQR = 3 mm). Furthermore, MDS is not based on approximations and does not need an iterative procedure to track sensor positions inside the implanted catheters. Conclusion Although both methods achieve similar results, MDS is to be preferred over rigid CPD while nonrigid CPD is unsuitable as it does not preserve topology.

Background Ultrasound is the first-line imaging modality for detection and classification of thyroid nodules. Certain characteristics observable by ultrasound have recently been identified that may indicate malignancy. This retrospective cohort study was conducted to test the hypothesis that advanced thyroid carcinomas show distinctive clinical and sonographic characteristics. Using a neural network model as proof of concept, nine clinical/sonographic features served as input. Methods All 96 study enrollees had histologically confirmed thyroid carcinomas, categorized (n = 32, each) as follows: group 1, advanced carcinoma (ADV) marked by local invasion or distant metastasis; group 2, non-advanced papillary carcinoma (PTC); or group 3, non-advanced follicular carcinoma (FTC). Preoperative ultrasound profiles were obtained via standardized protocols. The neural network had nine input neurons and one hidden layer. Results Mean age and the number of male patients in group 1 were significantly higher compared with groups 2 (p = 0.005) or 3 (p p p ≤ 0.01). Network accuracy in discriminating advanced vs. non-advanced tumors was 84.4% (95% confidence interval [CI]: 75.5–91), with positive and negative predictive values of 87.1% (95% CI: 70.2–96.4) and 92.3% (95% CI: 83.0–97.5), respectively. Conclusions Our study has shown some evidence that advanced thyroid tumors demonstrate distinctive clinical and sonographic characteristics. Further prospective investigations with larger numbers of patients and multicenter design should be carried out to show whether a neural network incorporating these features may be an asset, helping to classify malignancies of the thyroid gland.

A central question in neuroscience is how self-organizing dynamic interactions in the brain emerge on their relatively static structural backbone. Due to the complexity of spatial and temporal dependencies between different brain areas, fully comprehending the interplay between structure and function is still challenging and an area of intense research. In this paper we present a graph neural network (GNN) framework, to describe functional interactions based on the structural anatomical layout. A GNN allows us to process graph-structured spatio-temporal signals, providing a possibility to combine structural information derived from diffusion tensor imaging (DTI) with temporal neural activity profiles, like that observed in functional magnetic resonance imaging (fMRI). Moreover, dynamic interactions between different brain regions discovered by this data-driven approach can provide a multi-modal measure of causal connectivity strength. We assess the proposed model’s accuracy by evaluating its capabilities to replicate empirically observed neural activation profiles, and compare the performance to those of a vector auto regression (VAR), like that typically used in Granger causality. We show that GNNs are able to capture long-term dependencies in data and also computationally scale up to the analysis of large-scale networks. Finally we confirm that features learned by a GNN can generalize across MRI scanner types and acquisition protocols, by demonstrating that the performance on small datasets can be improved by pre-training the GNN on data from an earlier study. We conclude that the proposed multi-modal GNN framework can provide a novel perspective on the structure-function relationship in the brain. Accordingly this approach appears to be promising for the characterization of the information flow in brain networks.

It is generally assumed that the analysis of intracranial pressure (ICP) waveforms could be used for the detection of multiple cerebral pathophysiologies. A main obstacle for the analysis of ICP waveforms is given by the large variation of their generating signal, the arterial blood pressure (ABP). Using extended principal component analysis (PCA) we show that it is possible to distinguish between ICP waveforms generated by pathological ABP waveforms, e.g. in the case of a heart failure, without loosing information about the state of the cerebral compliance. We also create a dataset for ICP pulse classification that can be used to train models to distinguish between an intact and diminished cerebral compliance, as a function of the ICP-generating ABP pulse. As a baseline for classification we build a fully convolutional network (FCN) that reaches high performance with relatively few data.

Process monitoring and optimization in an industrial setting is a complex topic, a large amount of data is captured and its measurements are affected by many complex relations. There is, therefore, a growing necessity for methods to reduce its complexity and dimension such that it can be interpreted by humans. In this work we study the extraction of interpretable rules derived from components of a non-negative matrix factorization using real data.

High dose rate brachytherapy affords a frequent reassurance of the precise dwell positions of the radiation source. The current investigation proposes a multi-dimensional scaling transformation of both data sets to estimate dwell positions without any external reference. Furthermore, the related distributions of dwell positions are characterized by uni-or bi-modal heavy-tailed distributions. The latter are well represented by α-stable distributions. The newly proposed data analysis provides dwell position deviations with high accuracy, and, furthermore, offers a convenient visualization of the actual shapes of the catheters which guide the radiation source during the treatment.

Modern radiotherapy of female breast cancers often employs high dose rate brachytherapy, where a radioactive source is moved inside catheters, implanted in the female breast, according to a prescribed treatment plan. Source localization relative to the patient's anatomy is determined with solenoid sensors whose spatial positions are measured with an electromagnetic tracking system. Precise sensor dwell position determination is of utmost importance to assure irradiation of the cancerous tissue according to the treatment plan. We present a hybrid data analysis system which combines multi-dimensional scaling with particle filters to precisely determine sensor dwell positions in the catheters during subsequent radiation treatment sessions. Both techniques are complemented with empirical mode decomposition for the removal of superimposed breathing artifacts. We show that the hybrid model robustly and reliably determines the spatial positions of all catheters used during the treatment and precisely determines any deviations of actual sensor dwell positions from the treatment plan. The hybrid system only relies on sensor positions measured with an EMT system and relates them to the spatial positions of the implanted catheters as initially determined with a computed x-ray tomography.

This work presents an unsupervised mining strategy, applied to an independent component analysis (ICA) of segments of data collected while participants are answering to the items of the Halstead Category Test (HCT). This new methodology was developed to achieve signal components at trial level and therefore to study signal dynamics which are not available within participants’ ensemble average signals. The study will be focused on the signal component that can be elicited by the binary visual feedback which is part of the HCT protocol. The experimental study is conducted using a cohort of 58 participants.

This paper presents a convolutional neural network (CNN) which can be used for forecasting electricity load profiles 36 hours into the future. In contrast to well established CNN architectures, the input data is one-dimensional. A parameter scanning of network parameters is conducted in order to gain information about the influence of the kernel size, number of filters and number of nodes. Furthermore, different dropout methods are applied to the CNN and are evaluated. The results show that a good forecast quality can already be achieved with basic CNN architectures, the dropout improves the forecast. The method works not only for smooth sum loads of many hundred consumers, but also for the load of single apartment buildings.

Designing a multi-layer optical system with designated optical characteristics is an inverse design problem in which the resulting design is determined by several discrete and continuous parameters. In particular, we consider three design parameters to describe a multi-layer stack: Each layer's dielectric material and thickness as well as the total number of layers. Such a combination of both, discrete and continuous parameters is a challenging optimization problem that often requires a computationally expensive search for an optimal system design. Hence, most methods merely determine the optimal thicknesses of the system's layers. To incorporate layer material and the total number of layers as well, we propose a method that considers the stacking of consecutive layers as parameterized actions in a Markov decision process. We propose an exponentially transformed reward signal that eases policy optimization and adapt a recent variant of Q-learning for inverse design optimization. We demonstrate that our method outperforms human experts and a naive reinforcement learning algorithm concerning the achieved optical characteristics. Moreover, the learned Q-values contain information about the optical properties of multi-layer optical systems, thereby allowing physical interpretation or what-if analysis., Presented as a poster at the workshop on machine learning for engineering modeling, simulation and design @ NeurIPS 2020

Purpose: Currently methods for predicting absorbed dose after administering a radiopharmaceutical are rather crude in daily clinical practice. Most importantly individual tissue density distributions as well as local variations of the concentration of the radiopharmaceutical are commonly neglected. The current study proposes machine learning techniques like Green's function-based empirical mode decomposition and deep learning methods on U-net architectures in conjunction with soft tissue kernel Monte Carlo (MC) simulations to overcome current limitations in precision and reliability of dose estimations for clinical dosimetric applications. Methods: We present a hybrid method (DNN-EMD) based on deep neural networks (DNN) in combination with empirical mode decomposition (EMD) techniques. The algorithm receives Computed Tomography (CT) tissue density maps and dose maps, estimated according to the MIRD protocol, i.e., employing whole organ S-values and related time-integrated activities (TIAs), and from measured SPECT distributions of 177-Lu radionuclei, and learns to predict individual absorbed dose distributions. Density maps are replaced by their intrinsic modes as deduced from an EMD analysis. The system is trained using individual full MC simulations as reference. Results: Data from a patient cohort of 26 subjects are reported in this study. The proposed methods were validated employing a leave-one-out cross-validation technique. Deviations of estimated dose from corresponding MC results corroborate a superior performance of the newly proposed hybrid DNN-EMD method compared to its related MIRD DVK dose calculation. Not only are the mean deviations much smaller with the new method, but also the related variances are much reduced. If intrinsic modes of the tissue density maps are input to the algorithm, variances become even further reduced though the mean deviations are less affected. Conclusion: The newly proposed hybrid DNN-EMD method for individualized radiation dose prediction outperforms the MIRD DVK dose calculation method. It is fast enough to be of use in daily clinical practice.

In radionuclide therapies, dosimetry is used for determining patient-individual dose burden. Standard approaches provide whole organ doses only. For assessing dose heterogeneity inside organs, voxel-wise dosimetry based on 3D SPECT/CT imaging could be applied. Often, this is achieved by convolving voxel-wise time-activity-curves with appropriate dose-voxel-kernels (DVK). The DVKs are meant to model dose deposition, and can be more accurate if modelled for the specific tissue type under consideration. In literature, DVKs are often not adapted to these inhomogeneities, or simple approximation schemes are applied. For 26 patients, which had previously undergone a -PSMA or -DOTATOC therapy, decay maps, mass-density maps as well as tissue-type maps were derived from SPECT/CT acquisitions. These were used for a voxel-based dosimetry based on convolution with DVKs (each of size ) obtained by four different DVK methods proposed in literature. The simplest only considers a spatially constant soft-tissue DVK (herein named ‘constant’), while others either take into account only the local density of the center voxel of the DVK (herein named ‘center-voxel’) or scale each voxel linearly according to the proper mass density deduced from the CT image (herein named ‘density’) or considered both the local mass density as well as the direct path between the center voxel and any voxel in its surrounding (herein named ‘percentage’). Deviations between resulting dose values and those from full Monte-Carlo simulations (MC simulations) were compared for selected organs and tissue-types. For each DVK method, inter-patient variability was considerable showing both under- and over-estimation of energy dose compared to the MC result for all tissue densities higher than soft tissue. In kidneys and spleen, ‘constant’ and ‘density’-scaled DVKs achieved estimated doses with smallest deviations to the full MC gold standard (∼ underestimation). For low and high density tissue types such as lung and adipose or bone tissue, alternative DVK methods like ‘center-voxel’- and ‘percentage’- scaled achieved superior results, respectively. Concerning computational load, dose estimation with the DVK method ‘constant’ needs about 1.1 s per patient, center-voxel scaling amounts to 1.2 s, density scaling needs 1.4 s while percentage scaling consumes 860.3 s per patient. In this study encompassing a large patient cohort and four different DVK estimation methods, no single DVK-adaption method was consistently better than any other in case of soft tissue kernels. Hence in such cases the simplest DVK method, labeled ‘constant’, suffices. In case of tumors, often located in tissues of low (lung) or high (bone) density, more sophisticated DVK methods excel. The high inter-patient variability indicates that for evaluating new algorithms, a sufficiently large patient cohort needs to be involved.

The thrombus causing a stroke can be seen on the susceptibility weighted angiography (SWAN) magnetic resonance imaging (MRI) sequence. But it is very small and hard to detect by humans. Up to date the thrombus is identified by trained human experts. But as stroke needs quick treatment, an automatic detection of the thrombus would be useful to speed up the diagnosis of acute stroke. We propose a method for automatic thrombus detection from SWAN using three separate U-Nets which work on the axial, coronal and sagittal planes.

Independent component analysis (ICA), as a data driven method, has shown to be a powerful tool for functional magnetic resonance imaging (fMRI) data analysis. One drawback of this multivariate approach is, that it is naturally not convenient for analysis of group studies. Therefore various techniques have been proposed in order to overcome this limitation of ICA. In this paper a novel ICA based work-flow for extracting resting state networks from fMRI group studies is proposed. An empirical mode decomposition (EMD) is used to generate reference signals in a data driven manner, which can be incorporated into a constrained version of ICA (cICA), what helps to overcome the inherent ambiguities. The results of the proposed workflow are then compared to those obtained by a widely used group ICA approach. It is demonstrated that intrinsic modes, extracted by EMD, are suitable to serve as references for cICA to obtain typical resting state patterns, which are consistent over subjects. This novel processing pipeline makes it transparent for the user, how comparable activity patterns across subjects emerge, and also the trade-off between similarity across subjects and preserving individual features can be well adjusted and adapted for different requirements in the new work-flow.

In any radiotherapy, the absorbed dose needs to be estimated based on two factors, the time-integrated activity of the administered radiopharmaceutical and the patient-specific dose kernel. In this study, we consider the uncertainty with which such absorbed dose estimation can be achieved in a clinical environment. To calculate the total error of dose estimation we considered the following aspects: The error resulting from computing the time-integrated activity, the difference between the S-value and the patient specific full Monte Carlo simulation, the error from segmenting the volume-of-interest (kidney) and the intrinsic error of the activimeter. The total relative error in dose estimation can amount to 25.0% and is composed of the error of the time-integrated activity 17.1%, the error of the S-value 16.7%, the segmentation error 5.4% and the activimeter accuracy 5.0%. Errors from estimating the time-integrated activity and approximations applied to dose kernel computations contribute about equally and represent the dominant contributions far exceeding the contributions from VOI segmentation and activimeter accuracy.

We present a new method for decomposing two-dimensional data arrays with empirical mode decomposition (EMD). It performs envelope surface interpolation based on Green's functions in tension (GiT) to extract bi-dimensional intrinsic mode functions (BIMFs). The new method is called GiT-BEMD and outperforms existing bi-dimensional ensemble EMD (BEEMD) variants in terms of computational costs and quality of extracted intrinsic modes. More specifically, it is easy to implement, much faster than BEEMD, very robust and free from processing artifacts. GiT-BEMD is applied to fMRI data recorded during a contour integration task. Features extracted from resulting volume intrinsic mode functions (VIMFs) achieve higher classification accuracy compared to the canonical BEEMD. The new method thus provides a valuable alternative to existing mode decomposition methods for analyzing images.

Positron emission tomography (PET) provides a functional imaging modality to detect signs of dementias in human brains. Two-dimensional empirical mode decomposition (2D-EMD) provides means to analyze such images. It decomposes the latter into characteristic modes which represent textures on different spatial scales. These textures provide informative features for subsequent classification purposes. The study proposes a new EMD variant which relies on a Green's function based estimation method including a tension parameter to fast and reliably estimate the envelope hypersurfaces interpolating extremal points of the two-dimensional intensity distrubution of the images. The new method represents a fast and stable bi-dimensional EMD which speeds up computations roughly 100-fold. In combination with proper classifiers these exploratory feature extraction techniques can form a computer aided diagnosis (CAD) system to assist clinicians in identifying various diseases from functional images alone. PET images of subjects suffering from Alzheimer's disease are taken to illustrate this ability.

Background Empirical mode decomposition (EMD) is an empirical data decomposition technique. Recently there is growing interest in applying EMD in the biomedical field. New method EMDLAB is an extensible plug-in for the EEGLAB toolbox, which is an open software environment for electrophysiological data analysis. Results EMDLAB can be used to perform, easily and effectively, four common types of EMD: plain EMD, ensemble EMD (EEMD), weighted sliding EMD (wSEMD) and multivariate EMD (MEMD) on EEG data. In addition, EMDLAB is a user-friendly toolbox and closely implemented in the EEGLAB toolbox. Comparison with existing methods EMDLAB gains an advantage over other open-source toolboxes by exploiting the advantageous visualization capabilities of EEGLAB for extracted intrinsic mode functions (IMFs) and Event-Related Modes (ERMs) of the signal. Conclusions EMDLAB is a reliable, efficient, and automated solution for extracting and visualizing the extracted IMFs and ERMs by EMD algorithms in EEG study.

Digital museum guides - often together with eye trackers as innovative gadgets for intuitive interaction - provide attractive new ways for museums to communicate information to visitors and analyze their behaviour. In this paper, we investigate an approach to understand the gaze bedhaviour of persons viewing paintings in a museum. We present a method that can detect focussed areas (AOF) by analysing the fixation duration for the pixels of a painting. We can provide evidence that the viewing behaviour of laymen in a museum differs from what an expert expects according to the art historic relevance of certain regions of interest (ROI) in a painting. Consequently, museum educators have to apply intelligent assistance strategies that allow visitors to fully appreciate exhibits during their visit a of museum.

A novel Score-based Physarum Learner algorithm for learning Bayesian Network structure from data is introduced and shown to outperform common score based structure learning algorithms for some benchmark data sets. The Score-based Physarum Learner first initializes a fully connected Physarum-Maze with random conductances. In each Physarum Solver iteration, the source and sink nodes are changed randomly, and the conductances are updated. Connections exceeding a predefined conductance threshold are considered as Bayesian Network edges, and the score of the connected nodes are examined in both directions. A positive or negative feedback is given to the edge conductance based on the calculated scores. Due to randomness in selecting connections for evaluation, an ensemble of Score-based Physarum Learner is used to build the final Bayesian Network structure.

In cognitive neuroscience, extracting characteristic textures and features from functional imaging modalities which could be useful in identifying particular cognitive states across different conditions is still an important field of study. This paper explores the potential of two-dimensional ensemble empirical mode decomposition (2DEEMD) to extract such textures, so-called bidimensional intrinsic mode functions (BIMFs), of functional biomedical images, especially functional magnetic resonance images (fMRI) taken while performing a contour integration task. To identify most informative textures, i.e. BIMFs, a support vector machine (SVM) as well as a random forest (RF) classifier is trained for two different stimulus/response conditions. Classification performance is used to estimate the discriminative power of extracted BIMFs. The latter are then analyzed according to their spatial distribution of brain activations related with contour integration. Results distinctly show the participation of frontal brain areas in contour integration. Employing features generated from textures represented by BIMFs exhibit superior classification performance when compared with a canonical general linear model (GLM) analysis employing statistical parametric mapping (SPM).

NMF is a Blind Source Separation technique decomposing multivariate non-negative data sets into meaningful non-negative basis components and non-negative weights. In its canonical form an NMF algorithm was proposed by Lee and Seung (1999) [31] employing multiplicative update rules. In this study we show how the latter follow from a new variational Bayes NMF algorithm VBNMF employing a Gaussian noise kernel.

In this paper we propose a new method for volatile time series forecasting using Independent Component Analysis (ICA) algorithms and Savitzky-Golay filtering as preprocessing tools. The preprocessed data will be introduce in a based radial basis functions (RBF) Artificial Neural Network (ANN) and the prediction result will be compared with the one we get without these preprocessing tools or the classical Principal Component Analysis (PCA) tool.

In this work, a linear procedure to perform the intensity normalization of FP-CIT SPECT brain images is presented. This proposed methodology is based on the fact that the histogram of intensity values can be fitted accurately using a positive skewed α-stable distribution. Then, the predicted α-stable parameters and the location-scale property are used to linearly transform the intensity values in each voxel. This transformation is performed such that the new histograms in each image have a pre-specified α-stable distribution with desired location and dispersion values. The proposed methodology is compared with a similar approach assuming Gaussian distribution and the widely used specific-to-nonspecific ratio. In this work, we show that the linear normalization method using the α-stable distribution outperforms those existing methods.

A MRI brain image segmentation method using a hidden Markov random fields with heavy-tailed alpha-stable distributions is presented. Each brain tissue is modelled using an alpha-stable distribution. Then, a HMRF is used to include spatial information in the classification model. The Gaussian distribution has been widely used for the modelization of the cerebrospinal fluid, white matter and gray matter. Nevertheless, the alpha-stable distribution has been recently proposed as a more accurate alternative for this task. The alpha-stable distribution is more impulsive and is also able to model the asymmetry and heavy-tails of the histogram of the brain tissues. We have tested the proposed methodology in 18 MR images from the Internet Brain Segmentation Repository. The proposed methodology outperforms the segmentation results obtained when a Gaussian model for the histogram of the brain tissues is considered. Furthermore, as the Normal distribution is a particulaar case of alpha-stable distribution. Therefore, the proposed approach is also a generalization of the hidden Markov random field segmentation method with Gaussian distributions.

Background and objective The study follows the proposal of decomposing a given data matrix into a product of independent spatial and temporal component matrices. A multi-variate decomposition approach is presented, based on an approximate diagonalization of a set of matrices computed using a latent space representation. Methods The proposed methodology follows an algebraic approach, which is common to space, temporal or spatiotemporal blind source separation algorithms. More specifically, the algebraic approach relies on singular value decomposition techniques, which avoids computationally costly and numerically instable matrix inversion. The method is equally applicable to correlation matrices determined from second order correlations or by considering fourth order correlations. Results The resulting algorithms are applied to fMRI data sets either to extract the underlying fMRI components or to extract connectivity maps from resting state fMRI data collected for a dynamic functional connectivity analysis. Intriguingly, our algorithm shows increased spatial specificity compared to common approaches, while temporal precision stays similar. Conclusion The study presents a novel spatiotemporal blind source separation algorithm, which is both robust and avoids parameters that are difficult to fine tune. Applied on experimental data sets, the new method yields highly confined and focused areas with least spatial extent in the retinotopy case, and similar results in the dynamic functional connectivity analyses compared to other blind source separation algorithms. Therefore, we conclude that our novel algorithm is highly competitive and yields results, which are superior or at least similar to existing approaches.

International audience; The analysis of positron emission tomography (PET) scan image is challenging due to a high level of noise and a low resolution and also because differences between healthy and demented are very subtle. High dimensional classification methods based on PET have been proposed to automatically discriminate between normal control group (NC) patients and patients with Alzheimer’s disease (AD), with mild cognitive impairment (MCI), and mild cognitive impairment converting to Alzheimer’s disease (MCIAD) (a group of patients that clearly degrades to AD). We developed a voxelbased method for volumetric image analysis. We performed 3 classification experiments AD vs CG, AD vs MCI, MCIAD vs MCI. We will also give a small demonstration of the presented method on a set of face images. This method is capable to extract information about the location of metabolic changes induced by Alzheimer’s disease that directly relies statistical features and brain regions of interest (ROIs). We produce “maps” to visualize the most informative regions of the brain and compare them with voxel-wise statistics. Using the mean intensity of about 2000 6 × 6 × 6mm patches, selected by the extracted map, as input for a classifier we obtain a classification rate of 95.5%.

Objective: This paper explores the importance of the latent symmetry of the brain in computer-aided systems for diagnosing Alzheimer's disease (AD). Symmetry and asymmetry are studied from two points of view: (i) the development of an effective classifier within the scope of machine learning techniques, and (ii) the assessment of its relevance to the AD diagnosis in the early stages of the disease. Methods: The proposed methodology is based on eigenimage decomposition of single-photon emission-computed tomography images, using an eigenspace extension to accommodate odd and even eigenvectors separately. This feature extraction technique allows for support-vector-machine classification and image analysis. Results: Identification of AD patterns is improved when the latent symmetry of the brain is considered, with an estimated 92.78% accuracy (92.86% sensitivity, 92.68% specificity) using a linear kernel and a leave-one-out cross validation strategy. Also, asymmetries may be used to define a test for AD that is very specific (90.24% specificity) but not especially sensitive. Conclusions: Two main conclusions are derived from the analysis of the eigenimage spectrum. Firstly, the recognition of AD patterns is improved when considering only the symmetric part of the spectrum. Secondly, asymmetries in the hypo-metabolic patterns, when present, are more pronounced in subjects with AD.

The analysis of nonlinear and nonstationary time series is still a challenge, as most classical time series analysis techniques are restricted to data that is, at least, stationary. Empirical mode decomposition (EMD) in combination with a Hilbert spectral transform, together called Hilbert-Huang transform (HHT), alleviates this problem in a purely data-driven manner. EMD adaptively and locally decomposes such time series into a sum of oscillatory modes, called Intrinsic mode functions (IMF) and a nonstationary component called residuum. In this contribution, we propose an EMD-based method, called Sliding empirical mode decomposition (SEMD), which, with a reasonable computational effort, extends the application area of EMD to a true on-line analysis of time series comprising a huge amount of data if recorded with a high sampling rate. Using nonlinear and nonstationary toy data, we demonstrate the good performance of the proposed algorithm. We also show that the new method extracts component signals that fulfill all criteria of an IMF very well and that it exhibits excellent reconstruction quality. The method itself will be refined further by a weighted version, called weighted sliding empirical mode decomposition (wSEMD), which reduces the computational effort even more while preserving the reconstruction quality.

In this work we present a new method to reduce artifacts, produced by high-density objects, especially metal implants, in X-ray cone beam computed tomography (CBCT). These artifacts influence clinical diagnostics and treatments using CT data, if metal objects are located in the field of view (FOV). Our novel method reduces metal artifacts by virtually replacing the metal objects with tissue objects of the same shape. First, the considered objects must be segmented in the original 2D projection data as well as in a reconstructed 3D volume. The attenuation coefficients of the segmented voxels are replaced with adequate attenuation coefficients of tissue (or water), then the required parts of the volume are projected onto the segmented 2D pixels, to replace the original information. This corrected 2D data can then be reconstructed with reduced artifacts, i. e. all metal objects virtually vanished. After the reconstruction, the segmented 3D metal objects were re-inserted into the corrected 3D volume. Our method was developed for mobile C-arm CBCTs; as it is necessary that they are of low weight, the C-arm results in unpredictable distortion. This misalignment between the original 2D data and the forward projection of the reconstructed 3D volume must be adjusted before the correction of the segmented 2D pixels. We applied this technique to clinical data and will now present the results.

We propose a determinant criterion to constrain the solutions of non-negative matrix factorization problems and achieve unique and optimal solutions in a general setting, provided an exact solution exists. We demonstrate how optimal solutions are obtained by a heuristic named detNMF in an illustrative example and discuss the difference to sparsity constraints. Furthermore, an intuitive explanation of multi-layer techniques is discussed also.

In this work we present an approach to understand neuronal mechanisms underlying perceptual learning. Experimental results achieved with stimulus patterns of coherently moving dots are considered to build a simple neuronal model. The design of the model is made transparent and underlying behavioral assumptions made explicit. The key aspect of the suggested neuronal model is the learning algorithm used: We evaluated an implementation of Hebbian learning and are thus able to provide a straight-forward model capable to explain the neuronal dynamics underlying perceptual learning. Moreover, the simulation results suggest a very simple explanation for the aspect of "sub-threshold" learning (Watanabe et al. in Nature 413:844-884, 2001) as well as the relearning of motion discrimination after damage to primary visual cortex as recently reported (Huxlin et al. in J Neurosci 29:3981-3991, 2009) and at least indicate that perceptual learning might only occur when accompanied by conscious percepts.The online version of this article (doi:10.1007/s11571-010-9134-9) contains supplementary material, which is available to authorized users.

This letter presents a novel computer-aided diagnosis (CAD) technique for the early diagnosis of Alzheimer's disease (AD) based on non-negative matrix factorization (NMF) analysis applied to single photon emission computed tomography (SPECT) images. A baseline normalized SPECT database containing normalized data for both AD patients and healthy reference patients is selected for this study. The SPECT database is analyzed by applying the Fisher discriminant ratio (FDR) for feature selection and NMF for feature extraction of relevant components of each subject. The main goal of these preprocessing steps is to reduce the large dimensionality of the input data and to relieve the so called “curse of dimensionality” problem. The resulting NMF-transformed set of data, which contains a reduced number of features, is classified by means of a support vector machines based classification technique (SVM). The proposed NMF + SVM method yields up to 94% classification accuracy, with high sensitivity and specificity values (upper than 90%), becoming an accurate method for SPECT image classification. For the sake of completeness, comparison between another recently developed principal component analysis (PCA) plus SVM method and the proposed method is also provided, yielding results for the NMF + SVM approach that outperform the behavior of the reference PCA + SVM or conventional voxel-as-feature (VAF) plus SVM methods.

Nowadays, the accuracy of speech processing systems is strongly affected by acoustic noise. This is a serious obstacle regarding the demands of modern applications. Therefore, these systems often need a noise reduction algorithm working in combination with a precise voice activity detector (VAD). The computation needed to achieve denoising and speech detection must not exceed the limitations imposed by real time speech processing systems. This paper presents a novel VAD for improving speech detection robustness in noisy environments and the performance of speech recognition systems in real time applications. The algorithm is based on a Multivariate Complex Gaussian (MCG) observation model and defines an optimal likelihood ratio test (LRT) involving multiple and correlated observations (MCO) based on a jointly Gaussian probability distribution (jGpdf) and a symmetric covariance matrix. The complete derivation of the jGpdf-LRT for the general case of a symmetric covariance matrix is shown in terms of the Cholesky decomposition which allows to efficiently compute the VAD decision rule. An extensive analysis of the proposed methodology for a low dimensional observation model demonstrates: (i) the improved robustness of the proposed approach by means of a clear reduction of the classification error as the number of observations is increased, and (ii) the trade-off between the number of observations and the detection performance. The proposed strategy is also compared to different VAD methods including the G.729, AMR and AFE standards, as well as other recently reported algorithms showing a sustained advantage in speech/non-speech detection accuracy and speech recognition performance using the AURORA databases.

A novel method called binNMF is introduced which aimed to extract hidden information from multivariate binary data sets. The method treats the problem in the spirit of blind source separation: The data are assumed to be generated by a superposition of several simultaneously acting sources or elementary causes which are not observable directly. The superposition process is based on a minimum of assumptions and reversed to identify the underlying sources. The method is motivated, developed, and demonstrated in the context of binary wafer test data which evolve during microchip fabrication.

We present an automatic method for selecting regions of interest (ROIs) of the information contained in three-dimensional functional brain images using Gaussian mixture models (GMMs), where each Gaussian incorporates a contiguous brain region with similar activation. The novelty of the approach is based on approximating the grey-level distribution of a brain image by a sum of Gaussian functions, whose parameters are determined by a maximum likelihood criterion via the expectation maximization (EM) algorithm. Each Gaussian or cluster is represented by a multivariate Gaussian function with a center coordinate and a certain shape. This approach leads to a drastic compression of the information contained in the brain image and serves as a starting point for a variety of possible feature extraction methods for the diagnosis of brain diseases.