Data infrastructures for AI in medical imaging: a report on the experiences of five EU projects.

Artificial intelligence (AI) is transforming the field of medical imaging and has the potential to bring medicine from the era of 'sick-care' to the era of healthcare and prevention. The development of AI requires access to large, complete, and harmonized real-world datasets, representative of the population, and disease diversity. However, to date, efforts are fragmented, based on single-institution, size-limited, and annotation-limited datasets. Available public datasets (e.g., The Cancer Imaging Archive, TCIA, USA) are limited in scope, making model generalizability really difficult. In this direction, five European Union projects are currently working on the development of big data infrastructures that will enable European, ethically and General Data Protection Regulation-compliant, quality-controlled, cancer-related, medical imaging platforms, in which both large-scale data and AI algorithms will coexist. The vision is to create sustainable AI cloud-based platforms for the development, implementation, verification, and validation of trustable, usable, and reliable AI models for addressing specific unmet needs regarding cancer care provision. In this paper, we present an overview of the development efforts highlighting challenges and approaches selected providing valuable feedback to future attempts in the area.Key points• Artificial intelligence models for health imaging require access to large amounts of harmonized imaging data and metadata.• Main infrastructures adopted either collect centrally anonymized data or enable access to pseudonymized distributed data.• Developing a common data model for storing all relevant information is a challenge.• Trust of data providers in data sharing initiatives is essential.• An online European Union meta-tool-repository is a necessity minimizing effort duplication for the various projects in the area.

Sparsely Activated Networks.

Previous literature on unsupervised learning focused on designing structural priors with the aim of learning meaningful features. However, this was done without considering the description length of the learned representations, which is a direct and unbiased measure of the model complexity. In this article, first, we introduce the φ metric that evaluates unsupervised models based on their reconstruction accuracy and the degree of compression of their internal representations. We then present and define two activation functions [Identity and rectified linear unit (ReLU)] as a base of reference and three sparse activation functions (top-k absolutes, Extrema-Pool indices, and Extrema) as candidate structures that minimize the previously defined φ . We last present sparsely activated networks (SANs) that consist of kernels with shared weights that, during encoding, are convolved with the input and then passed through a sparse activation function. During decoding, the same weights are convolved with the sparse activation map, and subsequently, the partial reconstructions from each weight are summed to reconstruct the input. We compare SANs using the five previously defined activation functions on a variety of data sets (Physionet, UCI-epilepsy, MNIST, and FMNIST) and show that models that are selected using φ have small description representation length and consist of interpretable kernels.

Deep Learning in Cardiology.

The medical field is creating large amount of data that physicians are unable to decipher and use efficiently. Moreover, rule-based expert systems are inefficient in solving complicated medical tasks or for creating insights using big data. Deep learning has emerged as a more accurate and effective technology in a wide range of medical problems such as diagnosis, prediction, and intervention. Deep learning is a representation learning method that consists of layers that transform data nonlinearly, thus, revealing hierarchical relationships and structures. In this review, we survey deep learning application papers that use structured data, and signal and imaging modalities from cardiology. We discuss the advantages and limitations of applying deep learning in cardiology that also apply in medicine in general, while proposing certain directions as the most viable for clinical use.

Simulation of a Cellular Automaton with Markov Chains: Applications in Self-Organized Dynamical Systems.

over the past years, technology has allowed information technology to contemplate complex events as well as complex semantic features to predict what types of "thoughts" are being conceptualized. The introduction of the neuro-robotics field allows a mix of different disciplines to inter-collate and produce actual results that could be considered outputs of a science-fiction novel 20 twenty years ago. In the present work, we attempted to present an example of how an automaton can move in an environment with obstacles, by regulating its behavior so as to allow a decision based on rewards and penalties. Examples of the robotic behavior, running on a virtual environment are presented, along with a discussion of its different possibilities expressed as a penalty function for the behavior of the robot.

Signal2Image Modules in Deep Neural Networks for EEG Classification.

Deep learning has revolutionized computer vision utilizing the increased availability of big data and the power of parallel computational units such as graphical processing units. The vast majority of deep learning research is conducted using images as training data, however the biomedical domain is rich in physiological signals that are used for diagnosis and prediction problems. It is still an open research question how to best utilize signals to train deep neural networks.In this paper we define the term Signal2Image (S2Is) as trainable or non-trainable prefix modules that convert signals, such as Electroencephalography (EEG), to image-like representations making them suitable for training image-based deep neural networks defined as `base models'. We compare the accuracy and time performance of four S2Is (`signal as image', spectrogram, one and two layer Convolutional Neural Networks (CNNs)) combined with a set of `base models' (LeNet, AlexNet, VGGnet, ResNet, DenseNet) along with the depth-wise and 1D variations of the latter. We also provide empirical evidence that the one layer CNN S2I performs better in eleven out of fifteen tested models than non-trainable S2Is for classifying EEG signals and present visual comparisons of the outputs of some of the S2Is.

The "Prisoner's Dilemma" in the Tumor Biology Context: The Warburg Effect, Proliferation Dynamics and Equilibria.

There are numerous theories concerning carcinogenesis. Starting from the Warburg effect, which was one of the first theories concerning the mitochondrial dysfunction in tumor cells. Further on, the "two-hit" theory, where tumors were considered to be the outcome of genetic aberrations or mutations and more specifically of a certain number of "hits" each one resulting in a mutation. One of the main physical problems of biological systems is proliferation. Proliferation brings forwards two main questions: First, under a given population of cells, at time t what will be the precise population at time t+24h (or any other time point)? Second, what are the metabolic strategies followed by tumor cells in order to facilitate for their growth? In the present work we have used experimental data obtained from proliferation experiments of leukemic cells, where cell population and glucose consumption were evaluated. These data were further used to examine whether cells progress through competitive behavior or synergistically. Our results have shown that cells probably progress through a cooperative strategy.

Natural History of Carotid Atherosclerosis in Relation to the Hemodynamic Environment.

Carotid atherosclerosis may lead to devastating clinical outcomes such as stroke. Data on the value of local factors in predicting progression in carotid atherosclerosis are limited. Our aim was to investigate the association of local endothelial shear stress (ESS) and low-density lipoprotein (LDL) accumulation with the natural history of atherosclerotic disease using a series of 3 time points of human magnetic resonance data. Three-dimensional lumen/wall reconstruction was performed in 12 carotids, and blood flow and LDL mass transport modeling were performed. Our results showed that an increase in plaque thickness and a decrease in lumen size were associated with low ESS and high LDL accumulation in the arterial wall. Low ESS (odds ratio [OR]: 2.99; 95% confidence interval [CI]: 2.31-3.88; P < .001 vs higher ESS) and high LDL concentration (OR: 3.26; 95% CI: 2.44-4.36; P < .001 vs higher LDL concentration) were significantly associated with substantial local plaque growth. Low ESS and high LDL accumulation both presented a diagnostic accuracy of 67% for predicting plaque growth regions. Modeling of blood flow and LDL mass transport show promise in predicting progression of carotid atherosclerosis.

3-D registration on carotid artery imaging data: MRI for different timesteps.

A common problem which is faced by the researchers when dealing with arterial carotid imaging data is the registration of the geometrical structures between different imaging modalities or different timesteps. The use of the "Patient Position" DICOM field is not adequate to achieve accurate results due to the fact that the carotid artery is a relatively small structure and even imperceptible changes in patient position and/or direction make it difficult. While there is a wide range of simple/advanced registration techniques in the literature, there is a considerable number of studies which address the geometrical structure of the carotid artery without using any registration technique. On the other hand the existence of various registration techniques prohibits an objective comparison of the results using different registration techniques. In this paper we present a method for estimating the statistical significance that the choice of the registration technique has on the carotid geometry. One-Way Analysis of Variance (ANOVA) showed that the p-values were <;0.0001 for the distances of the lumen from the centerline for both right and left carotids of the patient case that was studied.

A proof-of-concept study for predicting the region of atherosclerotic plaque development based on plaque growth modeling in carotid arteries.

In this work, we present a computational model for plaque growth utilizing magnetic resonance data of a patient's carotid artery. More specifically, we model blood flow utilizing the Navier-Stokes equations, as well as LDL and HDL transport using the convection-diffusion equation in the arterial lumen. The accumulated LDL in the arterial wall is oxidized considering the protective effect of HDL. Macrophages recruitment and foam cells formation are the final step of the proposed multi-level modeling approach of the plaque growth. The simulated results of our model are compared with the follow-up MRI findings in 12 months regarding the change to the arterial wall thickness. WSS and LDL may indicate potential regions of plaque growth (R(2)=0.35), but the contribution of foam cells formation, macrophages and oxidized LDL increased the prediction significantly (R(2)=0.75).

Prediction of atheromatic plaque evolution in carotids using features extracted from the arterial geometry.

Knowing the arterial geometry might be helpful in the assessment of a plaque rupture event. We present a proof of concept study implementing a novel method which can predict the evolution in time of the atheromatic plaque in carotids using only statistical features which are extracted from the arterial geometry. Four feature selection methods were compared: Quadratic Programming Feature Selection (QPFS), Minimal Redundancy Maximal Relevance (mRMR), Mutual Information Quotient (MIQ) and Spectral Conditional Mutual Information (SPECCMI). The classifier used is the Support Vector Machines (SVM) with linear and Gaussian kernels. The maximum accuracy that was achieved in predicting the variation in the mean value of the Lumen distance from the centerline and the thickness was 71.2% and 70.7% respectively.

Fusion of optical coherence tomographic and angiographic data for more accurate evaluation of the endothelial shear stress patterns and neointimal distribution after bioresorbable scaffold implantation: comparison with intravascular ultrasound-derived reconstructions.

Intravascular ultrasound (IVUS)-based reconstructions have been traditionally used to examine the effect of endothelial shear stress (ESS) on neointimal formation. The aim of this analysis is to compare the association between ESS and neointimal thickness (NT) in models obtained by the fusion of optical coherence tomography (OCT) and coronary angiography and in the reconstructions derived by the integration of IVUS and coronary angiography. We analyzed data from six patients implanted with an Absorb bioresorbable vascular scaffold that had biplane angiography, IVUS and OCT investigation at baseline and 6 or 12 months follow-up. The IVUS and OCT follow-up data were fused separately with the angiographic data to reconstruct the luminal morphology at baseline and follow-up. Blood flow simulation was performed on the baseline reconstructions and the ESS was related to NT. In the OCT-based reconstructions the ESS were lower compared to the IVUS-based models (1.29 ± 0.66 vs. 1.87 ± 0.66 Pa, P = 0.030). An inverse correlation was noted between the logarithmic transformed ESS and the measured NT in all the OCT-based models which was higher than the correlation reported in five of the six IVUS-derived models (-0.52 ± 0.19 Pa vs. -0.10 ± 0.04, P = 0.028). Fusion of OCT and coronary angiography appears superior to IVUS-based reconstructions; therefore it should be the method of choice for the study of the effect of the ESS on neointimal proliferation.

An automatic electroencephalography blinking artefact detection and removal method based on template matching and ensemble empirical mode decomposition.

Electrooculographic (EOG) artefacts are one of the most common causes of Electroencephalogram (EEG) distortion. In this paper, we propose a method for EOG Blinking Artefacts (BAs) detection and removal from EEG. Normalized Correlation Coefficient (NCC), based on a predetermined BA template library was used for detecting the BA. Ensemble Empirical Mode Decomposition (EEMD) was applied to the contaminated region and a statistical algorithm determined which Intrinsic Mode Functions (IMFs) correspond to the BA. The proposed method was applied in simulated EEG signals, which were contaminated with artificially created EOG BAs, increasing the Signal-to-Error Ratio (SER) of the EEG Contaminated Region (CR) by 35 dB on average.