Chapter

U-Net Based Glioblastoma Segmentation with Patient’s Overall Survival Prediction

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Glioma is a type of malignant brain tumors which requires early detection for patients Overall Survival (OS) prediction and better treatment planning. This task can be simplified by computer aided automatic segmentation of brain MRI volumes into sub-regions. The MRI volumes segmentation can be achieved by deep learning methods but due to highly imbalance data, it becomes very challenging. In this article, we propose deep learning based solutions for Glioma segmentation and patient’s OS. To segment each pixel, we have designed a simplified version of 2D U-Net which is slice based and to predict OS, we have analyzed radiomic features. The training dataset of BraTS 2019 challenge is partitioned into train and test set and our primary results on test set are promising as dice score of (whole tumor 0.84, core tumor 0.80 and enhancing tumor 0.63) in glioma segmentation. Radiomic features based on intensity and shape are extracted from the MRI volumes and segmented tumor for OS prediction task. We further eliminate the low variance features using Recursive Features Elimination (RFE). The Random Forest Regression is used to predict OS time. By using intensities of peritumoral edema-label 2 of Flair, the necrotic and non-enhancing tumor core-label 1 along with enhancing tumor-label 4 of T1 contrast enhanced volumes and patients age, we are capable to predict patient’s OS with considerable accuracy of 31%.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... Several axial, sagittal, and coronal views were used to train the model to compensate for the absence of depth information in 2D models. A combination of 2D and lightweight 3D models was used to train the ensemble model introduced by Asra et al. [94] to extract both 2D and 3D benefits. The 3D dilated multi-fiber network (DMFNet) [53] model was selected because fewer parameters could be learnt more rapidly than other 3D complicated models. ...
... The segmentation classes are highly asymmetric. For a sample brain tumour MRI slice, about 98% of the voxels belong to either healthy tissue or the surrounding black region, 1.02% belong to the ED class, 0.29% belong to the ET class, and 0.23% belong to the NCR/NET class [94,126]. The class imbalance may be dealt with via appropriate training data sampling, better loss functions, and augmentation methods. ...
Article
Full-text available
Objective Glioblastoma multiforme (GBM) is a grade IV brain tumour with very low life expectancy. Physicians and oncologists urgently require automated techniques in clinics for brain tumour segmentation (BTS) and survival prediction (SP) of GBM patients to perform precise surgery followed by chemotherapy treatment. Methods This study aims at examining the recent methodologies developed using automated learning and radiomics to automate the process of SP. Automated techniques use pre-operative raw magnetic resonance imaging (MRI) scans and clinical data related to GBM patients. All SP methods submitted for the multimodal brain tumour segmentation (BraTS) challenge are examined to extract the generic workflow for SP. Results The maximum accuracies achieved by 21 state-of-the-art different SP techniques reviewed in this study are 65.5 and 61.7% using the validation and testing subsets of the BraTS dataset, respectively. The comparisons based on segmentation architectures, SP models, training parameters and hardware configurations have been made. Conclusion The limited accuracies achieved in the literature led us to review the various automated methodologies and evaluation metrics to find out the research gaps and other findings related to the survival prognosis of GBM patients so that these accuracies can be improved in future. Finally, the paper provides the most promising future research directions to improve the performance of automated SP techniques and increase their clinical relevance.
... To overcome the lack of depth information in 2D models, we train it in multiple views like axial, sagittal, and coronal. As 3D and 2D models have their respective advantages and to combine these advantages, we train our model with both 2D net-work by Asra et al. [10] and the lightweight 3D model known as Dilated Multi-Fiber Network (DMFNet) [12]. The main reason for selecting the 3D DMF model is that it has fewer number of parameters which can be trained quickly as compared to other 3D complex models. ...
... 2D Model. Motivated by the presentation of our previous 2D network [10], which performed better for HGG volumes, we used the same model for this segmentation. The slices were cropped to 224 × 224 in order to design a simpler 2D U-Net model. ...
Chapter
Full-text available
Automatic segmentation of gliomas from brain Magnetic Resonance Imaging (MRI) volumes is an essential step for tumor detection. Various 2D Convolutional Neural Network (2D-CNN) and its 3D variant, known as 3D-CNN based architectures, have been proposed in previous studies, which are used to capture contextual information. The 3D models capture depth information, making them an automatic choice for glioma segmentation from 3D MRI images. However, the 2D models can be trained in a relatively shorter time, making their parameter tuning relatively easier. Considering these facts, we tried to propose an ensemble of 2D and 3D models to utilize their respective benefits better. After segmentation, prediction of Overall Survival (OS) time was performed on segmented tumor sub-regions. For this task, multiple radiomic and image-based features were extracted from MRI volumes and segmented sub-regions. In this study, radiomic and image-based features were fused to predict the OS time of patients. Experimental results on BraTS 2020 testing dataset achieved a dice score of 0.79 on Enhancing Tumor (ET), 0.87 on Whole Tumor (WT), and 0.83 on Tumor Core (TC). For OS prediction task, results on BraTS 2020 testing leaderboard achieved an accuracy of 0.57, Mean Square Error (MSE) of 392,963.189, Median SE of 162,006.3, and Spearman R correlation score of −0.084.
... Furthermore, OS time prediction is based on radiomic features derived from a single view of an MRI image. [25][26][27] The OS time prediction schemes are unable to generalise well with the unseen data. When a slice is visualised in a single view (eg, axial), different neighbouring pixels in the region-of-interest (ROI) are observed as compared to the other two views (eg, sagittal and coronal). ...
... For the OS time prediction, accuracy, MSE, Median SE, SpearmanR, and stdSE metrics were used. 26 DSC was used to evaluate the segmentation model, which was needed to calculate the similarity between the model's predicted and actual labels. The average of these values was then computed to find the dice score, as shown in Equation (2) below. ...
Article
Full-text available
Glioblastoma (GBM) is the most high‐risk and grievous tumour in the brain that causes the death of more than 50% of the patients within one to 2 years after diagnosis. Accurate detection and prognosis of this disease are critical to provide essential guidelines for treatment planning. This study proposed using a deep learning‐based network for the GBM segmentation and radiomic features for the patient's overall survival (OS) time prediction. The segmentation model used in this study was a modified U‐Net‐based deep 3D multi‐level dilated convolutional neural network. It uses multiple kernels of altered sizes to capture contextual information at different levels. The proposed scheme for OS time prediction overcomes the problem of information loss caused by the derivation of features in a single view due to the variation in the neighbouring pixels of the tumorous region. The selected features were based on texture, shape, and volume. These features were computed from the segmented components of tumour in axial, coronal, and sagittal views of magnetic resonance imaging slices. The proposed models were trained and evaluated on the BraTS 2019 dataset. Experimental results of OS time prediction on the validation data have showed an accuracy of 48.3%, with the mean squared error of 92 599.598. On the validation data, the segmentation model achieved a mean dice similarity coefficient of 0.75, 0.89, and 0.80 for enhancing tumour, whole tumour, and tumour core, respectively. Future work is warranted to improve the overall performance of OS time prediction based on the findings in this study.
... Authors of [13][14][15] proposed the use of some handcrafted and radiomics features extracted from automatically segmented volumes with region labels to train a random forest regression model to predict the survival time of GBM patients in days. These studies achieved respectively 52%, 51.7%, and 27.25% accuracy on the validation set of the BraTS (Brain Tumor Image Segmentation) 2019 challenge. ...
Article
Full-text available
Glioma is the most common type of primary malignant brain tumor. Accurate survival time prediction for glioma patients may positively impact treatment planning. In this paper, we develop an automatic survival time prediction tool for glioblastoma patients along with an effective solution to the limited availability of annotated medical imaging datasets. Ensembles of snapshots of three dimensional (3D) deep convolutional neural networks (CNN) are applied to Magnetic Resonance Image (MRI) data to predict survival time of high-grade glioma patients. Additionally, multi-sequence MRI images were used to enhance survival prediction performance. A novel way to leverage the potential of ensembles to overcome the limitation of labeled medical image availability is shown. This new classification method separates glioblastoma patients into long- and short-term survivors. The BraTS (Brain Tumor Image Segmentation) 2019 training dataset was used in this work. Each patient case consisted of three MRI sequences (T1CE, T2, and FLAIR). Our training set contained 163 cases while the test set included 46 cases. The best known prediction accuracy of 74% for this type of problem was achieved on the unseen test set.
... Model/Methods used Brain tumor [106], [107], [110]- [112], [114]- [118] Base U-net [18], [109], [120] 3D U-net [81] Adversarial net; GAN [59], [108] Residual block [113] Dense block [87] Cascaded U-net [92] Residual block; Parallel U-net [44] Inception block; Up skip connections [45] Dense block; Inception block [119] 3D U-net; Residual block [19] 3D U-net, Inception block, Residual block Brain tissue [103], [121]- [124] Base U-net [28], [160] 3D U-net [161] 2.5D U-net [54] Residual block [101] Parallel U-net [41] Attention gate; Residual block White matter tracts [126], [127] U-net with modified skip connections [125] Base U-net [89] Cascaded U-net Fetal brain [128]- [130] Base U-net [131] Base U-net; 3D U-net Stroke lesion/thrombus [133]- [136] Base U-net [132] 3D U-net [69] Dense block; Inception block Cardiovascular structures [138], [140]- [142], [144], [146] Base U-net [62], [139], [143], [145] Residual block [4], [9], [10], [28] 3D U-net [86], [90] Cascaded U-net [5], [8] Cascaded 3D U-net [11] Base U-net; 3D U-net [83] Adversarial net; GAN [58] Residual block [137] Dense block [74] U-net++ [47] Inception block; Residual block [6] Cascaded 3D U-net; Residual block Prostate cancer [147], [149]- [151] Base U-net [28] 3D U-net [64] Recurrent net [58] Residual block Liver cancer [152], [153] Base U-net [21] 3D U-net Nasopharyngeal cancer [25] 3D U-net; Residual block [98] Parallel U-net [154] Modified convolution block Femur [12]- [14] 3D U-net Breast cancer [155] Base U-net [99] Parallel U-net Spinal cord [156], [157] Base U-net Blood vessels [100] Base U-net Placenta [158] Base U-net Uterus [159] Base U-net Vertebral column [17] 3D U-net ...
Preprint
Full-text available
U-net is an image segmentation technique developed primarily for medical image analysis that can precisely segment images using a scarce amount of training data. These traits provide U-net with a very high utility within the medical imaging community and have resulted in extensive adoption of U-net as the primary tool for segmentation tasks in medical imaging. The success of U-net is evident in its widespread use in all major image modalities from CT scans and MRI to X-rays and microscopy. Furthermore, while U-net is largely a segmentation tool, there have been instances of the use of U-net in other applications. As the potential of U-net is still increasing, in this review we look at the various developments that have been made in the U-net architecture and provide observations on recent trends. We examine the various innovations that have been made in deep learning and discuss how these tools facilitate U-net. Furthermore, we look at image modalities and application areas where U-net has been applied.
Article
Glioma is an aggressive type of cancer that develops in the brain or spinal cord. Due to many differences in its shape and appearance, accurate segmentation of glioma for identifying all parts of the tumor and its surrounding cancerous tissues is a challenging task. In recent researches, the combination of multi-atlas segmentation and machine learning methods provides robust and accurate results by learning from annotated atlas datasets. To overcome the side effects of limited existing information on atlas-based segmentation, and the long training phase of learning methods, we proposed a semi-supervised unified framework for multi-label segmentation that formulates this problem in terms of a Markov Random Field energy optimization on a parametric graph. To evaluate the proposed framework, we apply it to publicly available BRATS datasets, including low- and high-grade glioma tumors. Experimental results indicate competitive performance compared to the state-of-the-art methods. Compared with the top ranked methods, the proposed framework obtains the best dice score for segmenting of “whole tumor” (WT), “tumor core” (TC ) and “enhancing active tumor” (ET) regions. The achieved accuracy is 94\(\%\) characterized by the mean dice score. The motivation of using MRF graph is to map the segmentation problem to an optimization model in a graphical environment. Therefore, by defining perfect graph structure and optimum constraints and flows in the continuous max-flow model, the segmentation is performed precisely.
Article
Full-text available
The development of an automated glioma segmentation system from MRI volumes is a difficult task because of data imbalance problem. The ability of deep learning models to incorporate different layers for data representation assists medical experts like radiologists to recognize the condition of the patient and further make medical practices easier and automatic. State-of-the-art deep learning algorithms enable advancement in the medical image segmentation area, such a segmenting the volumes into sub-tumor classes. For this task, fully convolutional network (FCN)-based architectures are used to build end-to-end segmentation solutions. In this paper, we proposed a multi-level Kronecker convolutional neural network (MLKCNN) that captures information at different levels to have both local and global level contextual information. Our ML-KCNN uses Kronecker convolution, which overcomes the missing pixels problem by dilated convolution. Moreover, we used a post-processing technique to minimize false positive from segmented outputs, and the generalized dice loss (GDL) function handles the data-imbalance problem. Furthermore, the combination of connected component analysis (CCA) with conditional random fields (CRF) used as a post-processing technique achieves reduced Hausdorff distance (HD) score of 3.76 on enhancing tumor (ET), 4.88 on whole tumor (WT), and 5.85 on tumor core (TC). Dice similarity coefficient (DSC) of 0.74 on ET, 0.90 on WT, and 0.83 on TC. Qualitative and visual evaluation of our proposed method shown effectiveness of the proposed segmentation method can achieve performance that can compete with other brain tumor segmentation techniques.
Article
Full-text available
U-net is an image segmentation technique developed primarily for image segmentation tasks. These traits provide U-net with a high utility within the medical imaging community and have resulted in extensive adoption of U-net as the primary tool for segmentation tasks in medical imaging. The success of U-net is evident in its widespread use in nearly all major image modalities, from CT scans and MRI to Xrays and microscopy. Furthermore, while U-net is largely a segmentation tool, there have been instances of the use of U-net in other applications. Given that U-net’s potential is still increasing, this narrative literature review examines the numerous developments and breakthroughs in the U-net architecture and provides observations on recent trends. We also discuss the many innovations that have advanced in deep learning and discuss how these tools facilitate U-net. In addition, we review the different image modalities and application areas that have been enhanced by U-net.
Article
Glioma is a common type of tumor that develops in the brain. Due to many differences in the shape and appearance, accurate segmentation of glioma for identifying all parts of the tumor and its surrounding tissues in cancer detection is a challenging task in cancer detection. In recent researches, the combination of atlas-based segmentation and machine learning methods have presented superior performance over other automatic brain MRI segmentation algorithms. To overcome the side effects of limited existing information on atlas-based segmentation, and the long training and the time consuming phase of learning methods, we proposed a semi-supervised learning framework by introducing a probabilistic graph based method. It combines the advantages of label propagation and patch-based segmentation on a parametric graph. To evaluate the proposed framework, we apply it to publicly available BRATS datasets, including low and high-grade glioma tumors. The experimental results show that the proposed framework has accurate segmentation results. Compared with the state-of-the-art methods, the proposed framework could obtain the best dice score for segmenting the “whole tumor” (WT) and “tumor core” (TC) regions. The segmentation result of the “enhancing active tumor” (ET) region is similar to the most recent works compared in this paper.
Article
Full-text available
Objectives To identify and summarise all studies using statistical methods to project lung cancer incidence or mortality rates more than 5 years into the future. Study type Systematic review. Methods We performed a systematic literature search in multiple electronic databases to identify studies published from 1 January 1988 to 14 August 2018, which used statistical methods to project lung cancer incidence and/or mortality rates. Reference lists of relevant articles were checked for additional potentially relevant articles. We developed an organisational framework to classify methods into groups according to the type of data and the statistical models used. Included studies were critically appraised using prespecified criteria. Results One hundred and one studies met the inclusion criteria; six studies used more than one statistical method. The number of studies reporting statistical projections for lung cancer increased substantially over time. Eighty-eight studies used projection methods, which did not incorporate data on smoking in the population, and 16 studies used a method which did incorporate data on smoking. Age–period–cohort models (44 studies) were the most commonly used methods, followed by other generalised linear models (35 studies). The majority of models were developed using observed rates for more than 10 years and used data that were considered to be good quality. A quarter of studies provided comparisons of fitted and observed rates. While validation by withholding the most recent observed data from the model and then comparing the projected and observed rates for the most recent period provides important information on the model’s performance, only 12 studies reported doing this. Conclusion This systematic review provides an up-to-date summary of the statistical methods used in published lung cancer incidence or mortality projections. The assessment of the strengths of existing methods will help researchers to better apply and develop statistical methods for projecting lung cancer rates. Some of the common methods described in this review can be applied to the projection of rates for other cancer types or other non-infectious diseases.
Article
Full-text available
Gliomas are the most common primary brain malignancies. Accurate and robust tumor segmentation and prediction of patients' overall survival are important for diagnosis, treatment planning and risk factor identification. Here we present a deep learning-based framework for brain tumor segmentation and survival prediction in glioma, using multimodal MRI scans. For tumor segmentation, we use ensembles of three different 3D CNN architectures for robust performance through a majority rule. This approach can effectively reduce model bias and boost performance. For survival prediction, we extract 4,524 radiomic features from segmented tumor regions, then, a decision tree and cross validation are used to select potent features. Finally, a random forest model is trained to predict the overall survival of patients. The 2018 MICCAI Multimodal Brain Tumor Segmentation Challenge (BraTS), ranks our method at 2nd and 5th place out of 60+ participating teams for survival prediction tasks and segmentation tasks respectively, achieving a promising 61.0% accuracy on the classification of short-survivors, mid-survivors and long-survivors.
Article
Full-text available
In the Vehicular Ad-hoc NETworks (VANET), the collection and dissemination of life-threatening traffic event information by vehicles are of utmost importance. However, the traditional VANETs face several security issues. We propose a new type of blockchain to resolve critical message dissemination issues in the VANET. We create a local blockchain for real-world event message exchange among vehicles within the boundary of a country, which is a new type of blockchain suitable for the VANET. We present a public blockchain that stores the node trustworthiness and message trustworthiness in a distributed ledger that is appropriate for secure message dissemination.
Conference Paper
Full-text available
Machine and deep learning algorithms are rapidly growing in dynamic research of medical imaging. Currently, substantial efforts are developed for the enrichment of medical imaging applications using these algorithms to diagnose the errors in disease diagnostic systems which may result in extremely ambiguous medical treatments. Machine and deep learning algorithms are important ways in medical imaging to predict the symptoms of early disease. Deep learning techniques, in specific convolutional networks, have promptly developed a methodology of special for investigating medical images. It uses the supervised or unsupervised algorithms using some specific standard dataset to indicate the predictions. We survey image classification, object detection, pattern recognition, reasoning etc. concepts in medical imaging. These are used to improve the accuracy by extracting the meaningful patterns for the specific disease in medical imaging. These ways also indorse the decision-making procedure. The major aim of this survey is to highlight the machine learning and deep learning techniques used in medical images. We intended to provide an outline for researchers to know the existing techniques carried out for medical imaging, highlight the advantages and drawbacks of these algorithms, and to discuss the future directions. For the study of multi-dimensional medical data, machine and deep learning provide a commendable technique for creation of classification and automatic decision making. This paper provides a survey of medical imaging in the machine and deep learning methods to analyze distinctive diseases. It carries consideration concerning the suite of these algorithms which can be used for the investigation of diseases and automatic decision-making.
Article
Full-text available
Our research [Team Name: URAN] of 'Neuromorphic Neural Network for Multimodal Brain Tumor Segmentation and Survival analysis' demonstrates its performance on 'Overall Survival Prediction', without any prior training using multi-modal MRI Image segmentation by medical doctors (though provided by BraTS 2018 challenge data). Two segmentation categories are adopted instead of three segmentations , which is beyond the ordinary scope of BraTS 2018's segmentation challenge. The test results(the worst blind test, 51%: the nominal test, 71%) of 'survival prediction' challenge have proved the early feasibility of our neuromorphic neural network for helping the medical doctors, considering the human clinical accuracy of 50%~70%.
Chapter
Full-text available
Security is nonnegotiable key point for in-vehicle networking. However, all communication between Electrical Control Unites (ECU) still suffer from security drawbacks like highly processing time or preserving confidentiality, integrity and authenticity. In this paper, we propose an approach to assess the feasibility of a private Blockchain technology to overcome the aforementioned drawbacks. In this approach, we consider in-vehicle networking contains two parts, namely, central (or connected) gateway (cGW) and switches. cGW and switches are Blockchain nodes, wherein Blockchain consensus protocols are what keep all the nodes on a network synchronized with each other. The approach considers any communication type between ECUs as an individual event, which can be a transaction, data entry or application execution. A use case of secure communication between two ECUs is presented as an evaluation mechanism for securing in-vehicle networking using the proposed Blockchain approach.
Conference Paper
Full-text available
The article deals with optimization of input data for wireless IoT devices with database handled by cloud framework. Tested hardware consists of ESP32 board with Wi-Fi technology support (Thinger.IO cloud platform) and Sensor Tag development boards with integrated sensors and Bluetooth connection communicating with IBM Watson IoT framework. The new idea is how to send optimized measured data to IoT platform which is limited to one message per second usually. We propose two separate approaches for input data: accumulation data in IoT device to multi-value packet and noise elimination solved by advanced interpolation (Kalman filter). Both principles can be combined to reduce noise and increase of data frequency stored in cloud platform for next knowledge datamining. Data accumulation can be used for machine or product vibration analyses (acceleration data packed during 1 second) in cloud platform. Interpolated data can remove noise from actual product operational speed (gyroscope/magnetometer corrected value) to reduction false alarm messages in cloud framework.
Chapter
Full-text available
A cascade of fully convolutional neural networks is proposed to segment multi-modal Magnetic Resonance (MR) images with brain tumor into background and three hierarchical regions: whole tumor, tumor core and enhancing tumor core. The cascade is designed to decompose the multi-class segmentation problem into a sequence of three binary segmentation problems according to the subregion hierarchy. The whole tumor is segmented in the first step and the bounding box of the result is used for the tumor core segmentation in the second step. The enhancing tumor core is then segmented based on the bounding box of the tumor core segmentation result. Our networks consist of multiple layers of anisotropic and dilated convolution filters, and they are combined with multi-view fusion to reduce false positives. Residual connections and multi-scale predictions are employed in these networks to boost the segmentation performance. Experiments with BraTS 2017 validation set show that the proposed method achieved average Dice scores of 0.7859, 0.9050, 0.8378 for enhancing tumor core, whole tumor and tumor core, respectively. The corresponding values for BraTS 2017 testing set were 0.7831, 0.8739, and 0.7748, respectively.
Article
Full-text available
The model is difficult to establish because the principle of the locomotive adhesion process is complex. This paper presents a data-driven adhesion status fault diagnosis method based on deep learning theory. The adhesion coefficient and creep speed of a locomotive constitute the characteristic vector. The sparse autoencoder unsupervised learning network studies the input vector, and the single-layer network is superimposed to form a deep neural network. Finally, a small amount of labeled data is used to fine-tune training the entire deep neural network, and the locomotive adhesion state fault diagnosis model is established. Experimental results show that the proposed method can achieve a 99.3% locomotive adhesion state diagnosis accuracy and satisfy actual engineering monitoring requirements.
Article
Full-text available
There is no doubt that Internet of Things (IoT) occupy a very important role in our daily lives. Indeed, numerous objects that we use every time, are being equipped with electronic devices and protocol suites in order to make them interconnected and connected to the Internet. In IoT, things process and exchange data without human intervention. Therefore, because of this full autonomy, these entities need to recognize and authenticate each other as well as to ensure the integrity of their exchanged data. Otherwise, they will be the target of malicious users and malicious use. Due to the size and other features of IoT, it is almost impossible to create an efficient centralized authentication system. To remedy this limit, in this paper, we propose an original decentralized system called bubbles of trust, which ensures a robust identification and authentication of devices. Furthermore, it protects the data integrity and availability. To achieve such a goal, our approach relies on the security advantages provided by blockchains, and serves to create secure virtual zones (bubbles) where things can identify and trust each other. We also provided a real implementation1 of our mechanism using the C++ language and Ethereum blockchain. The obtained results prove its ability to satisfy IoT security requirements, its efficiency, and its low cost
Article
Full-text available
We propose a multiscale texture features based on Laplacian-of Gaussian (LoG) filter to predict progression free (PFS) and overall survival (OS) in patients newly diagnosed with glioblastoma (GBM). Experiments use the extracted features derived from 40 patients of GBM with T1-weighted imaging (T1-WI) and Fluid-attenuated inversion recovery (FLAIR) images that were segmented manually into areas of active tumor, necrosis, and edema. Multiscale texture features were extracted locally from each of these areas of interest using a LoG filter and the relation between features to OS and PFS was investigated using univariate (i.e., Spearman’s rank correlation coefficient, log-rank test and Kaplan-Meier estimator) and multivariate analyses (i.e., Random Forest classifier). Three and seven features were statistically correlated with PFS and OS, respectively, with absolute correlation values between 0.32 and 0.36 and p < 0.05. Three features derived from active tumor regions only were associated with OS (p < 0.05) with hazard ratios (HR) of 2.9, 3, and 3.24, respectively. Combined features showed an AUC value of 85.37 and 85.54% for predicting the PFS and OS of GBM patients, respectively, using the random forest (RF) classifier. We presented a multiscale texture features to characterize the GBM regions and predict he PFS and OS. The efficiency achievable suggests that this technique can be developed into a GBM MR analysis system suitable for clinical use after a thorough validation involving more patients.
Conference Paper
Full-text available
Increased use of Twitter by a large number of users has resulted in an unprecedented growth in the generation of user created data. This data is now being analyzed in various areas of research including that of user sentiment and behavior analysis during general elections. Rise in the use of Twitter for electoral campaigning has increased interest of researchers in this area. However while studies have utilized factors such as sentiment and frequency of tweets to evaluate candidate favorability, little research has been conducted in analyzing Twitter location data. This paper makes two major contributions. First, it studies the subjectivity and polarity of Twitter data based on the location of the tweets. Findings from this research will enable further fine grained analysis of Twitter data by using the location variable, allowing for more accurate results of user sentiment and behavior analysis. Second, we present a web-based system which can enable in the collection of Twitter data, extraction of the location, perform data analysis and then to plot them on a maps for visualization. We employ US presidential election data of 2016 as a case study to test our research questions and to demonstrate our results utilizing this web-based system.
Article
Full-text available
Vehicular networks enable vehicles to generate and broadcast messages in order to improve traffic safety and efficiency. However, due to the non-trusted environments, it is difficult for vehicles to evaluate the credibilities of received messages. In this paper, we propose a decentralized trust management system in vehicular networks based on blockchain techniques. In this system, vehicles can validate the received messages from neighboring vehicles using Bayesian Inference Model. Based on the validation result, the vehicle will generate a rating for each message source vehicle. With the ratings uploaded from vehicles, Roadside Units (RSUs) calculate the trust value offsets of involved vehicles and pack these data into a “block”. Then, each RSU will try to add their “blocks” to the trust blockchain which is maintained by all the RSUs. By employing the joint Proof-of-Work and Proof-of-Stake consensus mechanism, the more total value of offsets (stake) is in the block, the easier RSU can find the nonce for the hash function (proof-of-work). In this way, all RSUs collaboratively maintain an updated, reliable, and consistent trust blockchain. Simulation results reveal that the proposed system is effective and feasible in collecting, calculating, and storing trust values in vehicular networks.
Article
Full-text available
We present a novel and practical deep fully convolutional neural network architecture for semantic pixel-wise segmentation termed SegNet. This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer. The architecture of the encoder network is topologically identical to the 13 convolutional layers in the VGG16 network [1]. The role of the decoder network is to map the low resolution encoder feature maps to full input resolution feature maps for pixel-wise classification. The novelty of SegNet lies is in the manner in which the decoder upsamples its lower resolution input feature map(s). Specifically, the decoder uses pooling indices computed in the max-pooling step of the corresponding encoder to perform non-linear upsampling. This eliminates the need for learning to upsample. The upsampled maps are sparse and are then convolved with trainable filters to produce dense feature maps. We compare our proposed architecture with the widely adopted FCN [2] and also with the well known DeepLab-LargeFOV [3] , DeconvNet [4] architectures. This comparison reveals the memory versus accuracy trade-off involved in achieving good segmentation performance. SegNet was primarily motivated by scene understanding applications. Hence, it is designed to be efficient both in terms of memory and computational time during inference. It is also significantly smaller in the number of trainable parameters than other competing architectures and can be trained end-to-end using stochastic gradient descent. We also performed a controlled benchmark of SegNet and other architectures on both road scenes and SUN RGB-D indoor scene segmentation tasks. These quantitative assessments show that SegNet provides good performance with competitive inference time and most efficient inference memory-wise as compared to other architectures. We also provide a Caffe implementation of SegNet and a web demo at http://mi.eng.cam.ac.uk/projects/segnet/.
Article
Full-text available
Malaria remains a major burden on global health, with roughly 200 million cases worldwide and more than 400,000 deaths per year. Besides biomedical research and political efforts, modern information technology is playing a key role in many attempts at fighting the disease. One of the barriers toward a successful mortality reduction has been inadequate malaria diagnosis in particular. To improve diagnosis, image analysis software and machine learning methods have been used to quantify parasitemia in microscopic blood slides. This article gives an overview of these techniques and discusses the current developments in image analysis and machine learning for microscopic malaria diagnosis. We organize the different approaches published in the literature according to the techniques used for imaging, image preprocessing, parasite and cell segmentation, feature computation, and automatic cell classification. Readers will find the different techniques listed in tables, with the relevant articles cited next to them, for both thin and thick blood smear images. We also discussed the latest developments in sections devoted to deep learning and smartphone technology for future malaria diagnosis.
Article
Full-text available
In this paper, we report on the construction of a new framework for simulating mixed traffic consisting of cars, trams, and pedestrians that can be used to support discussions about road management, signal control, and public transit. Specifically, a layered road structure that was designed for car traffic simulations was extended to interact with an existing one-dimensional (1D) car-following model and a two-dimensional (2D) discrete choice model for pedestrians. The car model, pedestrian model, and interaction rules implemented in the proposed framework were verified through simulations involving simple road environments. The resulting simulated values were in near agreement with the empirical data. We then used the proposed framework to assess the impact of a tramway extension plan for a real city. The simulation results showed that the impact of the proposed tramway on existing car traffic would not be serious, and by extension, implied that the proposed framework could help stakeholders decide on expansion scenarios that are satisfactory to both tram users and private car owners.
Article
Full-text available
During the long history of computer vision, one of the grand challenges has been semantic segmentation which is the ability to segment an unknown image into different parts and objects (e.g., beach, ocean, sun, dog, swimmer). Furthermore, segmentation is even deeper than object recognition because recognition is not necessary for segmentation. Specifically, humans can perform image segmentation without even knowing what the objects are (for example, in satellite imagery or medical X-ray scans, there may be several objects which are unknown, but they can still be segmented within the image typically for further investigation). Performing segmentation without knowing the exact identity of all objects in the scene is an important part of our visual understanding process which can give us a powerful model to understand the world and also be used to improve or augment existing computer vision techniques. Herein this work, we review the field of semantic segmentation as pertaining to deep convolutional neural networks. We provide comprehensive coverage of the top approaches and summarize the strengths, weaknesses and major challenges.
Conference Paper
Full-text available
The focus of this study is to evaluate the impact of linguistic preprocessing and similarity functions for clustering Arabic Twitter tweets. The experiments apply an optimized version of the standard K-Means algorithm to assign tweets into positive and negative categories. The results show that root-based stemming has a significant advantage over light stemming in all settings. The Averaged Kullback-Leibler Divergence similarity function clearly outperforms the Cosine, Pearson Correlation, Jaccard Coefficient and Euclidean functions. The combination of the Averaged Kullback-Leibler Divergence and root-based stemming achieved the highest purity of 0.764 while the second-best purity was 0.719. These results are of importance as it is contrary to normal-sized documents where, in many information retrieval applications, light stemming performs better than root-based stemming and the Cosine function is commonly used.
Conference Paper
Full-text available
Deep clustering utilizes deep neural networks to learn feature representation that is suitable for clustering tasks. Though demonstrating promising performance in various applications, we observe that existing deep clustering algorithms either do not well take advantage of convolutional neural networks or do not considerably preserve the local structure of data generating distribution in the learned feature space. To address this issue, we propose a deep convolutional embedded clustering algorithm in this paper. Specifically, we develop a convolutional autoencoders structure to learn embedded features in an end-to-end way. Then, a clustering oriented loss is directly built on embedded features to jointly perform feature refinement and cluster assignment. To avoid feature space being distorted by the clustering loss, we keep the decoder remained which can preserve local structure of data in feature space. In sum, we simultaneously minimize the reconstruction loss of convolutional autoencoders and the clustering loss. The resultant optimization problem can be effectively solved by mini-batch stochastic gradient descent and back-propagation. Experiments on benchmark datasets empirically validate the power of convolutional autoencoders for feature learning and the effectiveness of local structure preservation.
Conference Paper
Full-text available
This paper discusses the Espressif Systems latest product ESP32 designed for Internet of Things and embedded system related projects. The ESP32 is a low-cost, low-power system on a chip series of microcontrollers with Wi-Fi and Bluetooth capabilities and a highly integrated structure powered by a dual-core Tensilica Xtensa LX6 microprocessor. This paper provides a comparative analysis of the ESP32 with some other market competitors and introduces the microcontroller specification, features and programming details. A portable, wireless oscilloscope based on the ESP-WROOM-32 and a mobile application is described in detail as an example of successful practical implementation of the device.
Conference Paper
Full-text available
Fully convolutional deep neural networks carry out excellent potential for fast and accurate image segmentation. One of the main challenges in training these networks is data imbalance, which is particularly problematic in medical imaging applications such as lesion segmentation where the number of lesion voxels is often much lower than the number of non-lesion voxels. Training with unbalanced data can lead to predictions that are severely biased towards high precision but low recall (sensitivity), which is undesired especially in medical applications where false negatives are much less tolerable than false positives. Several methods have been proposed to deal with this problem including balanced sampling, two step training, sample re-weighting, and similarity loss functions. In this paper, we propose a generalized loss function based on the Tversky index to address the issue of data imbalance and achieve much better trade-off between precision and recall in training 3D fully convolutional deep neural networks. Experimental results in multiple sclerosis lesion segmentation on magnetic resonance images show improved \(F_2\) score, Dice coefficient, and the area under the precision-recall curve in test data. Based on these results we suggest Tversky loss function as a generalized framework to effectively train deep neural networks.
Article
2018 Curran Associates Inc.All rights reserved. Batch Normalization (BatchNorm) is a widely adopted technique that enables faster and more stable training of deep neural networks (DNNs). Despite its pervasiveness, the exact reasons for BatchNorm's effectiveness are still poorly understood. The popular belief is that this effectiveness stems from controlling the change of the layers' input distributions during training to reduce the so-called “internal covariate shift”. In this work, we demonstrate that such distributional stability of layer inputs has little to do with the success of BatchNorm. Instead, we uncover a more fundamental impact of BatchNorm on the training process: it makes the optimization landscape significantly smoother. This smoothness induces a more predictive and stable behavior of the gradients, allowing for faster training.
Chapter
This paper proposes to use machine learning techniques with ultrasonic sensors to predict the behavior and status of a person when they live solely inside their house. The proposed system is tested on a single room. A grid of ultrasonic sensors is placed in the ceiling of a room to monitor the position and the status of a person (standing, sitting, lying down). The sensors readings are wirelessly communicated through a microcontroller to a cloud. An intelligent system will read the sensors values from the cloud and analyses them using machine learning algorithms to predict the person behavior and status and decide whether it is a normal situation or abnormal. If an abnormal situation is concluded, then an alert with be risen on a dashboard, where a care giver can take an immediate action. The proposed system managed to give results with accuracy exceeding 90%. Results out of this project will help people with supported needed, for example elderly people, to live their life as independent as possible, without too much interference from the caregivers. This will also free the care givers and allows them to monitors more units at the same time.
Article
Fully Convolutional Networks have been achieving remarkable results in image semantic segmentation, while being efficient. Such efficiency results from the capability of segmenting several voxels in a single forward pass. So, there is a direct spatial correspondence between a unit in a feature map and the voxel in the same location. In a convolutional layer, the kernel spans over all channels and extracts information from them. We observe that linear recombination of feature maps by increasing the number of channels followed by compression may enhance their discriminative power. Moreover, not all feature maps have the same relevance for the classes being predicted. In order to learn the inter-channel relationships and recalibrate the channels to suppress the less relevant ones, Squeeze and Excitation blocks were proposed in the context of image clas sification with Convolutional Neural Networks. However, this is not well adapted for segmentation with Fully Convolutional Networks since they segment several objects simultaneously, hence a feature map may contain relevant information only in some locations. In this paper, we propose recombination of features and a spatially adaptive recalibration block that is adapted for semantic segmentation with Fully Convolutional — Networks the SegSE block. Feature maps are recalibrated by considering the cross-channel information together with spatial relevance. Experimental results indicate that Recombination and Recalibration improve the results of a competitive baseline, and generalize across three different problems: brain tumor segmentation, stroke penumbra estimation, and ischemic stroke lesion outcome prediction. The obtained results are competitive or outperform the state of the art in the three applications.
Article
Glioblastoma multiforme (GBM)is a rapidly growing tumor associated with poor prognosis. This study evaluates the effectiveness of thirteen tumor shape features for overall survival (OS)prognosis in GBM patients. Shape features were extracted from the abnormality regions of the GBM tumor visible on the fluid attenuated inversion recovery (FLAIR)and T1-weighted contrast enhanced (T1CE)MR images of GBM patients. Survival analysis was performed using univariate and multivariate (with clinical features)Cox proportional hazards regression analysis. Kaplan-Meier survival curves were plotted and observed for the shape features which were found to be significant from the Cox regression analysis. Three 3D shape features: Bounding ellipsoid volume ratio (BEVR), sphericity and spherical disproportion, computed from both the abnormality regions were found to be significant for OS prognosis in GBM patients.
Article
Gliomas are the most frequent primary brain tumors, which have a high mortality. Surgery is the most commonly used treatment. Magnetic resonance imaging (MRI)is especially useful to assess gliomas and evaluate the success of the treatment in clinical practice. So accurately segmenting the brain tumor from MRI images is the key to clinical diagnostics and treatment planning. However, a large quantity of data produced by MRI prevents manual segmentation in a reasonable time. So automatic approaches are required for quick and effective segmentation. But the spatial and structural variability among brain tumors bring a challenge to automatically segment MRI images. Deep Convolutional Neural Network (DCNN)can achieve complex function mapping. This contributes to response to the challenge. Therefore, DCNN is the commonly used method in brain tumor segmentation. Although a large number of DCNN based methods have been proposed, these methods mainly aim to improve the quality of the image features extracted by DCNN. Actually, these methods usually ignore the prior knowledge in medical images, and the simple prior knowledge in brain tumor segmentation is that most of the tumor regions in images are left-right asymmetry. Based on this, in this paper, a novel deep convolutional neural network which combines symmetry have been proposed to automatically segment brain tumors. Our neural networks, called Deep Convolutional Symmetric Neural Network (DCSNN), extends DCNN based segmentation networks by adding symmetric masks in several layers. Our proposal was validated in the BRATS 2015 database, and we also give some baselines. The results are evaluated by dice similarity coefficient metric (DSC). And our proposed method achieved a competitive result with average DSC of 0.852. Furthermore, our method only takes about 10.8 s to segment a patient case. Although our method is not the best performance in BRATS 2015 challenge, as far as we know, our method outperforms other recent DCNN-based methods.
Article
We present an efficient deep learning approach for the challenging task of tumor segmentation in multisequence MR images. In recent years, Convolutional Neural Networks (CNN) have achieved state-of-the-art performances in a large variety of recognition tasks in medical imaging. Because of the considerable computational cost of CNNs, large volumes such as MRI are typically processed by subvolumes, for instance slices (axial, coronal, sagittal) or small 3D patches. In this paper we introduce a CNN-based model which efficiently combines the advantages of the short-range 3D context and the long-range 2D context. Furthermore, we propose a network architecture with modality-specific subnetworks in order to be more robust to the problem of missing MR sequences during the training phase. To overcome the limitations of specific choices of neural network architectures, we describe a hierarchical decision process to combine outputs of several segmentation models. Finally, a simple and efficient algorithm for training large CNN models is introduced. We evaluate our method on the public benchmark of the BRATS 2017 challenge on the task of multiclass segmentation of malignant brain tumors. Our method achieves good performances and produces accurate segmentations with median Dice scores of 0.918 (whole tumor), 0.883 (tumor core) and 0.854 (enhancing core).
Article
In the Internet of Vehicles (IoV), data sharing among vehicles is critical to improving driving safety and enhancing vehicular services. To ensure security and traceability of data sharing, existing studies utilize efficient Delegated Proof-of-Stake consensus scheme as hard security solutions to establish blockchainenabled IoV (BIoV). However, as the miners are selected from miner candidates by stake-based voting, defending against voting collusion between the candidates and compromised high-stake vehicles becomes challenging. To address the challenge, in this paper, we propose a two-stage soft security enhancement solution: (i) miner selection and (ii) block verification. In the first stage, we design a reputation-based voting scheme to ensure secure miner selection. This scheme evaluates candidates' reputation using both past interactions and recommended opinions from other vehicles. The candidates with high reputation are selected to be active miners and standby miners. In the second stage, to prevent internal collusion among active miners, a newly generated block is further verified and audited by standby miners. To incentivize the participation of the standby miners in block verification, we adopt the contract theory to model the interactions between active miners and standby miners, where block verification security and delay are taken into consideration. Numerical results based on a real-world dataset confirm the security and efficiency of our schemes for data sharing in BIoV.
Conference Paper
Determining the position of a body is crucial in the context of the Ambient Assisted Living: knowing if a person is standing, lying down or sitting makes the difference in the process of detecting anomalies due to inactivity. We present a non-invasive monitoring system to recognize if a person is seated using Bluetooth Low Energy technology. The system is composed by devices that can be easily attached to the furniture: a transmitter module that continuously broadcasts Bluetooth signal, and a receiver which collects the RSSi value of the received messages. Finally, a central server evaluates the attenuation of the Bluetooth signal within the 2.4GHz band, correlating it to the presence of a human body in the proximity of the device.
Conference Paper
Optical Character Recognition (OCR) can provide tele-measurement for legacy utility meters (water, natural gas, electricity etc) and can be accepted as a low-cost extension for the capabilities of these counters and for their time of life. On the other side, the “galvanic separation” provided by optical acquisition of displayed counter values offers the confidence of secure reading - for natural gas meters and similar instruments. Our cost-effective solution enables simple snapshots of counters displays (with simple devices like an “ArdtrCarn” - camera for Arduino embedded solutions), local “edge computing” for image packing (in. BMP format) and tele-transmission via the WiFi hot-spot of the intelligent building. At the server-side, National Instruments software sub-systems - NI Vision Development Module and LabView- provide OCR and cloud aggregation of measured values. Our solution is related to a research project for cloud-based smart metering (the LabVIEW sub-system belongs to the communications work-package). Specific to our contribution is the re-balancing of local and centralized processing, involving an assisted learning phase for the OCR.
Article
Glioblastoma multiforme (GBM) are aggressive brain tumors, which lead to poor overall survival (OS) of patients. OS prediction of GBM patients provides useful information for surgical and treatment planning. Radiomics research attempts at predicting disease prognosis, thus providing beneficial information for personalized treatment from a variety of imaging features extracted from multiple MR images. In this study, MR image derived texture features, tumor shape and volumetric features, and patient age were obtained for 163 patients. OS group prediction was performed for both 2-class (short and long) and 3-class (short, medium and long) survival groups. Support vector machine classification based recursive feature elimination method was used to perform feature selection. The performance of the classification model was assessed using 5-fold cross-validation. The 2-class and 3-class OS group prediction accuracy obtained were 98.7% and 88.95% respectively. The shape features used in this work have been evaluated for OS prediction of GBM patients for the first time. The feature selection and prediction scheme implemented in this study yielded high accuracy for both 2-class and 3-class OS group predictions. This study was performed using routinely acquired MR images for GBM patients, thus making the translation of this work into a clinical setup convenient.
Chapter
Leishmaniasis is considered a neglected disease that causes thousands of deaths annually in some tropical and subtropical countries. There are various techniques to diagnose leishmaniasis of which manual microscopy is considered to be the gold standard. There is a need for the development of automatic techniques that are able to detect parasites in a robust and unsupervised manner. In this paper we present a procedure for automatizing the detection process based on a deep learning approach. We train a U-net model that successfully segments leismania parasites and classifies them into promastigotes, amastigotes and adhered parasites.
Article
Blockchain can potentially be deployed in a wide range of applications due to its capability to ensure decentralisation, transparency and immutability. In this paper, we design a cryptographic membership authentication scheme (i.e. authenticating graph data) to support blockchain-based identity management systems (BIMS). Such a system is designed to bind a digital identity object to its real-world entity. Specifically, we introduce a new Transitively Closed Undirected Graph Authentication (TCUGA) scheme, which only needs to use node signatures (e.g. certificates for identifying nodes). The trapdoor hash function used in our scheme allows the signer to efficiently update the certificates without the need to re-sign the nodes. In other words, our scheme is efficient even though the graph dynamically adds or deletes vertices and edges. Moreover, our proposal can efficiently provide a proof when the edge between two vertices does not exist; thus, solving the existing intractability issue in transitive signature (the main tool for authenticating graph data). Finally, we prove the security of our proposed TCUGA in the standard model and evaluate its performance to show its feasibility for BIMS.
Article
Deep learning is a promising approach for extracting accurate information from raw sensor data from IoT devices deployed in complex environments. Because of its multilayer structure, deep learning is also appropriate for the edge computing environment. Therefore, in this article, we first introduce deep learning for IoTs into the edge computing environment. Since existing edge nodes have limited processing capability, we also design a novel offloading strategy to optimize the performance of IoT deep learning applications with edge computing. In the performance evaluation, we test the performance of executing multiple deep learning tasks in an edge computing environment with our strategy. The evaluation results show that our method outperforms other optimization solutions on deep learning for IoT.
Article
Exploiting social media data by extracting key information from it is one of the great challenges in data mining and knowledge discovery. Every election campaign has an online presence of voters which uses these social media platform to express their sentiments and opinions towards political parties, leaders and important topics. This paper presents a novel data collection technique for prediction of election outcomes and a topic modeling method for extracting topics. Data collection technique used RSS (Rich Site Summary) feeds of news articles and trending keywords from Twitter simultaneously and constructed an intelligent prediction model based primarily on the volume of tweets and sentiment of users. This paper effort to improve electoral predictions using social media data based dynamic keyword methodology. Different techniques for electoral prediction based on social media data has been investigated based on existing literature and isolate the factors which improve our methodology. Meaningful inferences such as the popularity of leaders and parties during different intervals, trending issues, and important factors are extracted from the data set. The election outcomes are compared with traditional methods used by survey agencies for exit polls and validation of results showed that social media data can predict with better accuracy. The research has identified that data collection technique and timing play an important role in yielding better accuracy in predicting outcomes and extracting meaningful inferences.
Article
Social media now plays a pivotal role in electoral campaigns. Rapid dissemination of information through platforms such as Twitter has enabled politicians to broadcast their message to a wide audience. In this paper, we investigated the sentiment of tweets by the two main presidential candidates, Hillary Clinton and Donald Trump, along with almost 2.9 million tweets by Twitter users during the 2016 US Presidential Elections. We analyzed these short texts to evaluate how accurately Twitter represented the public opinion and real world events of significance related with the elections. We also analyzed the behavior of over a million distinct Twitter users to identify whether the platform was used to share original opinions and to interact with other users or whether few opinions were repeated over and over again with little inter-user dialogue. Finally, we wanted to assess the sentiment of tweets by both candidates and their impact on the election related discourse on Twitter. Some of our findings included the discovery that little original content was created by users and Twitter was primarily used for rebroadcasting already present opinions in the form of retweets with little communication between users. Also of significance was the finding that sentiment and topics expressed on Twitter can be a good proxy of public opinion and important election related events. Moreover, we found that Donald Trump offered a more optimistic and positive campaign message than Hillary Clinton and enjoyed better sentiment when mentioned in messages by Twitter users.
Article
In recent years, underwater video technologies allow us to explore the ocean in scientific and noninvasive ways, such as environmental monitoring, marine ecology studies, and fisheries management. However the low-light and high-noise scenarios pose great challenges for the underwater image and video analysis. We here propose a CNN knowledge transfer framework for underwater object recognition and tackle the problem of extracting discriminative features from relatively low contrast images. Even with the insufficient training set, the transfer framework can well learn a recognition model for the special underwater object recognition task together with the help of data augmentation. For better identifying objects from an underwater video, a weighted probabilities decision mechanism is introduced to identify the object from a series of frames. The proposed framework can be implemented for real-time underwater object recognition on autonomous underwater vehicles and video monitoring systems. To verify the effectiveness of our method, experiments on a public dataset are carried out. The results show that the proposed method achieves promising results for underwater object recognition on both test image datasets and underwater videos.
Conference Paper
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif- ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implemen- tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called dropout that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry
Conference Paper
Ambient Intelligence is growing phenomena caused by advances of speed and size of computational hardware. It is possible to collect data from various sensors, devices or services and react to evaluated event and start the predefined process. These systems are closely connected with Cloud based services as they provide publicly available interface for management and visualization, computational and storage capabilities for forecasts and other advanced data processing to create ambient intelligence environment. The true ambient environments react to the presence of people and the sensors and actuators have become part of the environment.