Deep learning segmentation of Primary Sjögren's syndrome affected salivary glands from ultrasonography images.

Salivary gland ultrasonography (SGUS) has proven to be a promising tool for diagnosing various diseases manifesting with abnormalities in salivary glands (SGs), including primary Sjögren's syndrome (pSS). At present, the major obstacle for establishing SUGS as a standardized tool for pSS diagnosis is its low inter/intra observer reliability. The aim of this study was to address this problem by proposing a robust deep learning-based solution for the automated segmentation of SGUS images. For these purposes, four architectures were considered: a fully convolutional neural network, fully convolutional "DenseNets" (FCN-DenseNet) network, U-Net, and LinkNet. During the course of the study, the growing HarmonicSS cohort included 1184 annotated SGUS images. Accordingly, the algorithms were trained using a transfer learning approach. With regard to the intersection-over-union (IoU), the top-performing FCN-DenseNet (IoU = 0.85) network showed a considerable margin above the inter-observer agreement (IoU = 0.76) and slightly above the intra-observer agreement (IoU = 0.84) between clinical experts. Considering its accuracy and speed (24.5 frames per second), it was concluded that the FCN-DenseNet could have wider applications in clinical practice. Further work on the topic will consider the integration of methods for pSS scoring, with the end goal of establishing SGUS as an effective noninvasive pSS diagnostic tool. To aid this progress, we created inference (frozen models) files for the developed models, and made them publicly available.

Machine Learning Approach for Predicting Wall Shear Distribution for Abdominal Aortic Aneurysm and Carotid Bifurcation Models.

Computer simulations based on the finite element method represent powerful tools for modeling blood flow through arteries. However, due to its computational complexity, this approach may be inappropriate when results are needed quickly. In order to reduce computational time, in this paper, we proposed an alternative machine learning based approach for calculation of wall shear stress (WSS) distribution, which may play an important role in mechanisms related to initiation and development of atherosclerosis. In order to capture relationships between geometric parameters, blood density, dynamic viscosity and velocity, and WSS distribution of geometrically parameterized abdominal aortic aneurysm (AAA) and carotid bifurcation models, we proposed multivariate linear regression, multilayer perceptron neural network and Gaussian conditional random fields (GCRF). Results obtained in this paper show that machine learning approaches can successfully predict WSS distribution at different cardiac cycle time points. Even though all proposed methods showed high potential for WSS prediction, GCRF achieved the highest coefficient of determination (0.930-0.948 for AAA model and 0.946-0.954 for carotid bifurcation model) demonstrating benefits of accounting for spatial correlation. The proposed approach can be used as an alternative method for real time calculation of WSS distribution.

Minimum redundancy maximum relevance feature selection approach for temporal gene expression data.

BACKGROUND: Feature selection, aiming to identify a subset of features among a possibly large set of features that are relevant for predicting a response, is an important preprocessing step in machine learning. In gene expression studies this is not a trivial task for several reasons, including potential temporal character of data. However, most feature selection approaches developed for microarray data cannot handle multivariate temporal data without previous data flattening, which results in loss of temporal information. We propose a temporal minimum redundancy - maximum relevance (TMRMR) feature selection approach, which is able to handle multivariate temporal data without previous data flattening. In the proposed approach we compute relevance of a gene by averaging F-statistic values calculated across individual time steps, and we compute redundancy between genes by using a dynamical time warping approach. RESULTS: The proposed method is evaluated on three temporal gene expression datasets from human viral challenge studies. Obtained results show that the proposed method outperforms alternatives widely used in gene expression studies. In particular, the proposed method achieved improvement in accuracy in 34 out of 54 experiments, while the other methods outperformed it in no more than 4 experiments. CONCLUSION: We developed a filter-based feature selection method for temporal gene expression data based on maximum relevance and minimum redundancy criteria. The proposed method incorporates temporal information by combining relevance, which is calculated as an average F-statistic value across different time steps, with redundancy, which is calculated by employing dynamical time warping approach. As evident in our experiments, incorporating the temporal information into the feature selection process leads to selection of more discriminative features.

Real-time monitoring of cytotoxic effects of electroporation on breast and colon cancer cell lines.

PURPOSE: To study the effects of electroporation on different cell lines. MATERIAL: The effects of electroporation on human breast cancer (MDA-MB-231), human colon cancer (SW-480 and HCT-116), human fibroblast cell line (MRC-5), primary human aortic smooth muscle cells (hAoSMC) and human umbilical vein endothelial cells (HUVEC) were studied. Real-time technology was used for cell viability monitoring. Acridine orange/ethidium bromide assay was applied for cell death type determination. A numerical model of electroporation has been proposed. RESULTS: Electroporation induced inhibition of cell viability on dose (voltage) dependent way. The electroporation treatment 375-437.5Vcm-1 caused irreversible electroporation of cancer cells and reversible electroporation of healthy cells. The application of lower voltage rating (250Vcm-1) led to apoptosis as the predominant type of cell death, whereas the use of higher voltage (500Vcm-1) mainly caused necrosis. CONCLUSION: Electroporation represents a promising method in cancer treatment. Different cancer cell lines had different response to the identical electroporation treatment. Electroporation 375-437.5Vcm-1 selectively caused permanent damage of cancer cells (SW-480), while healthy cells (MRC-5, hAoSM and HUVEC) recovered after 72h. The type of cell death is dependent of electroporation conditions. The proposed numerical model is useful for the analysis of phenomena related to electroporation treatment.

Automated development of artificial neural networks for clinical purposes: Application for predicting the outcome of choledocholithiasis surgery.

Among various expert systems (ES), Artificial Neural Network (ANN) has shown to be suitable for the diagnosis of concurrent common bile duct stones (CBDS) in patients undergoing elective cholecystectomy. However, their application in practice remains limited since the development of ANNs represents a slow process that requires additional expertize from potential users. The aim of this study was to propose an ES for automated development of ANNs and validate its performances on the problem of prediction of CBDS. Automated development of the ANN was achieved by applying the evolutionary assembling approach, which assumes optimal configuring of the ANN parameters by using Genetic algorithm. Automated selection of optimal features for the ANN training was performed using a Backward sequential feature selection algorithm. The assessment of the developed ANN included the evaluation of predictive ability and clinical utility. For these purposes, we collected data from 303 patients who underwent surgery in the period from 2008 to 2014. The results showed that the total bilirubin, alanine aminotransferase, common bile duct diameter, number of stones, size of the smallest calculus, biliary colic, acute cholecystitis and pancreatitis had the best prognostic value of CBDS. Compared to the alternative approaches, the ANN obtained by the proposed ES had better sensitivity and clinical utility, which are considered to be the most important for the particular problem. Besides the fact that it enabled the development of ANNs with better performances, the proposed ES significantly reduced the complexity of ANNs' development compared to previous studies that required manual selection of optimal features and/or ANN configuration. Therefore, it is concluded that the proposed ES represents a robust and user-friendly framework that, apart from the prediction of CBDS, could advance and simplify the application of ANNs for solving a wider range of problems.

Parameter optimization of a computer-aided diagnosis system for detection of masses on digitized mammograms.

BACKGROUND: Reading mammograms is a difficult task and for this reason any development that may improve the performance in breast cancer screening is of great importance. OBJECTIVE: We proposed optimized computer aided diagnosis (CAD) system, equipped with reliability estimate module, for mass detection on digitized mammograms. METHODS: Proposed CAD system consists of four major steps: preprocessing, segmentation, feature extraction and classification. We propose a simple regression function as a threshold function for extraction of potential masses. By running optimization procedure we estimate parameters of the preprocessing and segmentation steps thus ensuring maximum mass detection sensitivity. In addition to the classification, where we tested seven different classifiers, the CAD system is equipped with reliability estimate module. RESULTS: By performing segmentation 91.3% of masses were correctly segmented with 4.14 false positives per image (FPpi). This result is improved in the classification phase where, among the seven tested classifiers, multilayer perceptron neural network achieved the best result including 77.4% sensitivity and 0.49 FPpi. CONCLUSION: By using the proposed regression function and parameter optimization we were able to improve segmentation results comparing to the literature. In addition, we showed that CAD system has high potential for being equipped with reliability estimate module.

Prediction of coronary plaque location on arteries having myocardial bridge, using finite element models.

This study was performed to evaluate the influences of the myocardial bridges on the plaque initializations and progression in the coronary arteries. The wall structure is changed due to the plaque presence, which could be the reason for multiple heart malfunctions. Using simplified parametric finite element model (FE model) of the coronary artery having myocardial bridge and analyzing different mechanical parameters from blood circulation through the artery (wall shear stress, oscillatory shear index, residence time), we investigated the prediction of "the best" position for plaque progression. We chose six patients from the angiography records and used data from DICOM images to generate FE models with our software tools for FE preprocessing, solving and post-processing. We found a good correlation between real positions of the plaque and the ones that we predicted to develop at the proximal part of the myocardial bridges with wall shear stress, oscillatory shear index and residence time. This computer model could be additional predictive tool for everyday clinical examination of the patient with myocardial bridge.

Computer simulation of three-dimensional plaque formation and progression in the carotid artery.

Atherosclerosis is becoming the number one cause of death worldwide. In this study, three-dimensional computer model of plaque formation and development for human carotid artery is developed. The three-dimensional blood flow is described by the Navier-Stokes equation, together with the continuity equation. Mass transfer within the blood lumen and through the arterial wall is coupled with the blood flow and is modeled by a convection-diffusion equation. The low-density lipoproteins transports in lumen of the vessel and through the vessel tissue are coupled by Kedem-Katchalsky equations. The inflammatory process is modeled using three additional reaction-diffusion partial differential equations. Fluid-structure interaction is used to estimate effective wall stress analysis. Plaque growth functions for volume progression are correlated with shear stress and effective wall stress distribution. We choose two specific patients from MRI study with significant plaque progression. Plaque volume progression using three time points for baseline, 3- and 12-month follow up is fitted. Our results for plaque localization correspond to low shear stress zone and we fitted parameters from our model using nonlinear least-square method. Determination of plaque location and composition, and computer simulation of progression in time for a specific patient shows a potential benefit for the prediction of disease progression. The proof of validity of three-dimensional computer modeling in the evaluation of atherosclerotic plaque burden may shift the clinical information of MRI from morphological assessment toward a functional tool. Understanding and prediction of the evolution of atherosclerotic plaques either into vulnerable or stable plaques are major tasks for the medical community.

Mining data from hemodynamic simulations for generating prediction and explanation models.

One of the most common causes of human death is stroke, which can be caused by carotid bifurcation stenosis. In our work, we aim at proposing a prototype of a medical expert system that could significantly aid medical experts to detect hemodynamic abnormalities (increased artery wall shear stress). Based on the acquired simulated data, we apply several methodologies for1) predicting magnitudes and locations of maximum wall shear stress in the artery, 2) estimating reliability of computed predictions, and 3) providing user-friendly explanation of the model's decision. The obtained results indicate that the evaluated methodologies can provide a useful tool for the given problem domain.