Search

Optical coherence tomography (OCT) is an imaging modality that is used extensively for ophthalmic diagnosis, near-histological visualization, and quantification of retinal abnormalities such as cysts, exudates, retinal layer disorganization, etc. Intra-retinal cysts (IRCs) occur in several macular disorders such as, diabetic macular edema, retinal vascular disorders, age-related macular degeneration, and inflammatory disorders. Automated segmentation of IRCs poses challenges owing to variations in the acquisition system scan intensities, speckle noise, and imaging artifacts. Several segmentation methods have been proposed in the literature for IRC segmentation on vendor-specific OCT images that lack generalizability across imaging systems. In this paper, we propose a fully convolutional network (FCN) model for vendor-independent IRC segmentation. The proposed method counteracts image noise variabilities and trains FCN models on OCT sub-images from the OPTIMA cyst segmentation challenge dataset (with four different vendor-specific images, namely, Cirrus, Nidek, Spectralis, and Topcon). Further, optimal data augmentation and model hyperparametrization are shown to prevent over-fitting for IRC area segmentation. The proposed method is evaluated on the test dataset with a recall/precision rate of 0.66/0.79 across imaging vendors. The Dice correlation coefficient of the proposed method outperforms that of the published algorithms in the OPTIMA cyst segmentation challenge with a Dice rate of 0.71 across the vendors.

Saliency based sentence ranking is a basic step of extractive text summarization. Saliency of a sentence is often measured based on the important words that the sentence contains. One of the drawbacks of such saliency-based sentence extraction method is that it extracts mainly the sentences related to the most common topic in the document. But the input document may contain multiple topics or events and the users may like to see in the summary the salient information for each different topic or event. To alleviate such problem, diversity-based re-ranking approach or sentence clustering-based approach is commonly used. But re-ranking or sentence clustering makes the summarization process slow. In this paper, we propose a novel summarization method that computes the score of a sentence by combining saliency and novelty of the sentence. Without using any re-ranker or clustering of sentences, the proposed approach can automatically take care of the diversity issue while producing a summary. We have evaluated the performance of the system on DUC 2001 and DUC 2002 benchmark single document summarization datasets. Our experiments reveal that it outperforms several existing state-of-the-art extractive summarization approaches.

This paper describes an approach to generic Bengali text summarization using latent semantic analysis (LSA). Our proposed LSA based single document summarization method uses the latent semantic analysis technique to identify semantically important sentences. We have compared our proposed approach with some state-of-the art summarization approaches. The results demonstrate that our proposed Bengali text summarization approach is effective.

Prediction of epidemics such as Foot and Mouth Disease (FMD) is a global necessity in addressing economic, political and ethical issues faced by the affected countries. In the absence of precise and accurate spatial information regarding disease dynamics, learning- based predictive models can be used to mimic latent spatial parameters so as to predict the spread of epidemics in time. This paper analyzes temporal predictions from four such learning-based models, namely: neural network, autoregressive, Bayesian network, and Monte-Carlo simulation models. The prediction qualities of these models have been validated using FMD incidence reports in Turkey. Additionally, the authors perform simulations of mitigation strategies based on the predictive models to curb the impact of the epidemic. This paper also analyzes the cost-effectiveness of these mitigation strategies to conclude that vaccinations and movement ban strategies are more cost-effective than premise culls before the onset of an epidemic outbreak; however, in the event of existing epidemic outbreaks, premise culling is more effective at controlling FMD.

Reoccurring instances of Foot and Mouth Disease (FMD) in countries with underdeveloped infection surveillance have been taking a heavy toll on the lives of millions of livestock and causing annual financial losses worth billions of US dollars every year. FMD has been widely analyzed as a highly infectious disease that spreads rapidly with the movement of infected or contaminated animals of multiple species, and movement of people as well. Here, a metapopulation based stochastic model is implemented to assess the FMD infection dynamics and to curb the economic losses in such countries. This model predicts the spatio-temporal evolution of FMD over a weighted contact network where the weights are characterized by the effect of wind and movement of animals and humans. FMD Incidence data from Turkey is used to calibrate and validate the model, and the predictive performance of our model is compared with baseline models as well. Finally, cost effective mitigation strategies are simulated using the theoretical concept of network fragmentation. Based on the theoretical reduction in the total number of infected animals, several practical mitigation strategies are proposed and their cost effectivenesses are also analyzed.

Prediction of epidemics such as Foot and Mouth Disease (FMD) is a global necessity in addressing economic, political and ethical issues faced by the affected countries. In the absence of precise and accurate spatial information regarding disease dynamics, learning- based predictive models can be used to mimic latent spatial parameters so as to predict the spread of epidemics in time. This paper analyzes temporal predictions from four such learning-based models, namely: neural network, autoregressive, Bayesian network, and Monte-Carlo simulation models. The prediction qualities of these models have been validated using FMD incidence reports in Turkey. Additionally, the authors perform simulations of mitigation strategies based on the predictive models to curb the impact of the epidemic. This paper also analyzes the cost-effectiveness of these mitigation strategies to conclude that vaccinations and movement ban strategies are more cost-effective than premise culls before the onset of an epidemic outbreak; however, in the event of existing epidemic outbreaks, premise culling is more effective at controlling FMD.

Currently many commercial buildings are not continuously monitored for energy consumption, especially small buildings which constitute 90% of all such buildings. However, readily available data from the electric meters can be used for monitoring and analyzing energy consumption. Efficient utilization of available historical data (from these meters) can potentially improve energy efficiency, help to identify common energy wasting problems, and, in the future, enable various Smart Grid programs, such as demand response, real-time pricing etc. This paper describes application of computational intelligence techniques for prediction of electric power consumption. The proposed approach combines regression and clustering methods, in order to improve the prediction accuracy of power consumption, as a function of time (of the day) and temperature, using real-life data from several commercial and government buildings. Empirical comparisons show that the proposed approach provides an improvement over the currently used bin-based method for modeling power consumption.

Containing an epidemic at its origin is the most desirable mitigation. Epidemics have often originated in rural areas, with rural communities among the first affected. Disease dynamics in rural regions have received limited attention, and results of general studies cannot be directly applied since population densities and human mobility factors are very different in rural regions from those in cities. We create a network model of a rural community in Kansas, USA, by collecting data on the contact patterns and computing rates of contact among a sampled population. We model the impact of different mitigation strategies detecting closely connected groups of people and frequently visited locations. Within those groups and locations, we compare the effectiveness of random and targeted vaccinations using a Susceptible-Exposed-Infected-Recovered compartmental model on the contact network. Our simulations show that the targeted vaccinations of only 10% of the sampled population reduced the size of the epidemic by 34.5%. Additionally, if 10% of the population visiting one of the most popular locations is randomly vaccinated, the epidemic size is reduced by 19%. Our results suggest a new implementation of a highly effective strategy for targeted vaccinations through the use of popular locations in rural communities., 17 pages, 4 figures