Bessam Abdulrazak AmI Lab, Università di Sherbrooke, Canada Hamdi Aloulou Università di Monastir, Centro per la ricerca digitale,. Franco Mercalli MultiMed Engineers SRLS, Italia Fulvio Mastrogiovanni Università di Genova, Italia Hisato Kobayashi Hosei University, Giappone.
Alzheimer’s Disease Early Detection Using a Low Cost Three-Dimensional
Densenet-121 Architecture
1 Introduction
This work consists in measuring the accuracy of the detection of Alzheimer's disease in a three-dimensional CNN architecture, specifically a densenet-121, trained using ADNI MRI images. Before presenting the results of the development of a low-cost sealing net for the detection of Alzheimer's disease, we first provide in Sect.2 some background definitions to support our work.
2 Background
Clinical Disease Stages
We can design DNNs to integrate them into computer-aided diagnosis protocols for the detection of many priority diseases. The early detection of AD with software will allow us to strengthen and improve medical protocols by providing what we call Computer Aided Diagnosis (CAD).
Medical Imaging
Positrons react with electrons in the body and when these two particles combine, they cancel each other out. Detectors in a PET scanner measure these photons and use this information to create images of internal organs [16].
3 Previous Work
This annihilation produces a small amount of energy in the form of two photons that shoot off in opposite directions. Finally, we want our process to be repeatable, and we report it complete with all the parameters used, as explained in the next sections.
4 Methodology
Data Acquisition
In fact, we did not find any article with multiclass classification with more than four classes. Finally, the quantitative analysis of the collected items does not make a great contribution due to these shortcomings.
Data Preprocessing
Our Development
With this implementation, we configure the neural network training process with the following parameters. Channels We send parameter 1 to the neural network constructor because the images are monochrome.
5 Results and Discussion
For example, it is common to test or validate neural networks during training; thus, it is possible to analyze the loss and accuracy of neural networks in each period. As the area under each curve approaches the value of 1.0, the classifier's diagnostic ability is shown to be greater.
6 Conclusions and Future Work
Similarly, LMCI's bad count is also acceptable because the class is usually classified as AD. However, as can be noted, the macro average and the weighted average are better.
We propose an adaptive configuration of the vital signs monitoring process depending on the variation in the patient's health status and the decisions of the medical staff. It provides a manual and self-adaptive configuration of the vital signs monitoring process depending on the variation in the patient's health status and the decisions of the medical staff.
Machine Learning Based Rank Attack Detection for Smart Hospital
Infrastructure
DIO (DODAG Information Object): used to build, maintain DODAG and regularly refresh the information of nodes in the network topology. A rank attack is one of the most well-known RPL attacks, where an attacker advertises a false rank to trick other nodes into establishing routes through him.
2 Related Work
3 Rank Attack Scenario
ML-based Rank Attack Detection for Smart Hospital Infrastructure 33 attacks are considered one of the dangerous attacks in dynamic IoT networks as the attacker controls an existing node (one of the internal attacks that can affect the RPL) in the DODAG or it can identify the network and insert its own malicious node and that node will act as the attack node as shown in Fig.4.
4 Proposed Approach
It mainly performs some extremely complex data transformations to find a solution to separate the data based on defined labels or outputs. The concept of the SVM learning approach is based on the definition of the optimal separating hyperplane (Figure 5) [21], which maximizes the limit of the training data [17,18].
5 IDS Solution and Results
Simulation Setup
In two-dimensional space this hyper-lane is a line that divides a plane into two parts where each class lies on either side.
Evaluation Parameter
LPM Power consumption parameter that indicates the power consumption in sleep mode. CPU flow parameter indicating the level of node processing. Send parameter related to node communication while transmitting Listen parameter related to node communication while receiving.
Power Tracking per Mote for Each Simulation
6 Conclusion
Winter, T., Thubert, P., Brandt, A., et al.: RPL: IPv6 routing protocol for low-power, lossy networks. Rehman, A., et al.: Rank attack using objective function in RPL for low power and lossy networks.
Remote Health Monitoring Systems Based on Bluetooth Low Energy (BLE)
Communication Systems
General Context
Therefore, various communication technologies were suggested for data exchange between body sensor nodes and the coordinator (first level). IEEE is the standard dedicated to the communication between the sensors and the coordinator.
Contributions and Paper Organisation
Today, sensing devices with 802.15.6 modules are not sufficiently available for commercial use in the market and are more expensive than Bluetooth Low Energy (BLE) wearable sensors, which are widely commercialized by many sensor manufacturers such as libelium, mindwave. For this reason, our research in this paper focuses on RHMS using sensors with BLE interfaces.
2 BLE Based RHMS
RHMS Basic Architecture
Related Works About RHMS Using BLE
According to the authors, a peer-reviewed evaluation of the developed monitoring system for 40 individuals (aged 18 to 66 years) showed that the proposed system is convenient and reliable and generates warning messages to the physician and patient in critical circumstances. The data from each sensor is transmitted to the storage server via the BLE communication interface.
3 BLE Communication Protocols
Basic Concepts
The master coordinates the MAC using a TDMA scheme, determines when slaves should listen, and provides them with the map of data channels to use.
BLE Protocols Stack
4 Reading Sensed Physiological Signs with BLE
Example of Services and Characteristics of HDP
Steps for Reading Sensed Data on Mobile App
5 Conclusion
Alfian, G., Syafrudin, M., Ijaz, M.F., Syaekhoni, M.A., Fitriyani, N.L., Rhee, J.: A customized healthcare monitoring system for diabetic patients using BLE-based sensors and data processing in real time. Kakria, P., Tripathi, N.K., Kitipawang, P.: A real-time human health monitoring system for remote cardiac patients using smartphones and wearable sensors.
Modeling and Specification
Most of these patterns are presented visually and informally, with no formal semantics involved. We propose a graphical modeling of these patterns to describe both their structural and behavioral features.
2 Structural Patterns Modeling
Metamodel
Registration design patterns” allow to register the attributes and the characteristics of a new device on the back-end server. Registration Design Patterns' are composed of 'Automatic Client Driven Registration Pattern' and 'Server Driven Model Pattern'.
General Pattern Model
Bootstrap's Medium-Based Model” allows you to configure a new device in-place via a removable storage medium inserted into the device. If the device is deployed locally, we use the Medium-Based Bootstrap Template as a solution to configure the new device.
3 Behavioral Patterns Modeling
4 Case Study: Smart Home
5 Patterns Specification
The ExtendedComponent diagram, which models the structural features of the design patterns, is transformed into a context in the Event-B method, in which we specify the architecture entities and their relationships. The sequence diagram is transformed into a machine in event B, in which we specify the events between the pattern entities.
6 Tool Support
This transformation is proposed to assign formal notations to IoT design patterns for the purpose of verifying their design correctness in the second step. This strategy is interesting because it defines the pattern development process and improves the quality of the obtained models and thus the success of the formal development process.
7 Related Work
For pattern specification, we used the Event-B formal method to assign formal notations to SOA design patterns in order to verify their design correctness. In this paper, we present the modeling of IoT design patterns proposed by Reinfurt et al.
8 Conclusions
Tounsi, I., Hadj Kacem, M., Hadj Kacem, A., Drira, K.: A Refinement-Based Approach for Building Valid SOA Design Patterns. Tounsi, I., Hrichi, Z., Hadj Kacem, M., Hadj Kacem, A., Drira, K.: Using SoaML Models and Event-B Specifications for Modeling SOA Design Patterns.
EEG-Based Hypo-vigilance Detection Using Convolutional Neural Network
In this paper, we focus on the EEG signal study recorded by fourteen electrodes for hypo-vigilance detection by analyzing the different functionalities in the brain from the electrodes placed on the participant's scalp. In this paper, we propose a CNN hypo-vigilance detection method using EEG data to classify drowsiness and wakefulness states.
2 Proposed Approach
Data Acquisition
The state of relaxation was characterized by alpha waves having a frequency interval between 8 to 12 Hz and an amplitude interval between 20 to 60 µV. The state of drowsiness was shown by theta waves, which have a frequency interval between 4 and 8 Hz and an amplitude interval between 50 and 75 µV.
Data Analysis: Simple CNN Classification
To reduce the overlap and increase the testing accuracy, we use the data augmentation technique [17] which consists in increasing the training set by data transformations that preserve the labels.
3 Experimental Evaluation
For our data distribution, we choose 70% for the train part and 30% for the test. After convergence, the optimal number of test epochs for all results of different electrodes set a value equal to 80.
4 Conclusion
The images or other third-party material in this chapter are included in the chapter's Creative Commons license, unless otherwise noted in a credit line for the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by law or exceeds the permitted use, you must obtain permission directly from the copyright holder.
Respiratory Activity Classification Based on Ballistocardiogram Analysis
As a consequence, it is recommended to have continuous monitoring of the vital signs to ensure an optimal diagnosis of a patient's condition [6]. Due to the ejection of the blood during systole, the body's mechanical reaction and thus the BCG signal are measured.
2 Material and Method
- Data Collection
- BCG Signal Analysis
- BCG Signal Feature Extraction
- Activities Classification
- Classification Evaluation
Both signals reach the upper limit of the acquisition equipment, which is explained by the broad peaks in the BCG signals. BCG Feature Engineering: The SFM and SC measurements are evaluated on each frame of the BCG signal.
3 Experimental Results
Feature Illustration
The terms of the confusion matrix presented in Figure 4 are defined as follows: where C is the number of classes, Me is the number of predictions of class i that actually belong to class j, this is usually measured by comparing the test results with the ground truth. Both coughing and movement activities (right plots) show large fluctuations in their respective sfm signals, this is due to the absence of periodicity in these signals.
Classifier Evaluation
We used a reconstructed time series signal from the spectral flatness measure (SFM) and the spectral centroid (SC) of the raw data. In: Proceedings of the 16th ACM International Symposium on Mobile Ad Hoc Networking and Computing, MobiHoc 2015, New York, NY, USA, pp.
A Convolutional Neural Network for Lentigo Diagnosis
Lentigo Detection
Most forms of lentigo are benign [13], such as lentigo simplex as Figure 1(a) and solar lentigo as Figure 1(b). For all these reasons, our approach is based on images obtained by this method.
Convolutional Neural Networks
3 Proposed Method for Lentigo Detection
- Data Preparation
- InceptionV3 Model
- Transfer Learning
- Prediction Model
In the first step of the preprocessing procedure, the RCM images from the training set are adjusted to match the InceptionV3 network. As shown in Fig.3, transfer learning [22] is proposed to ensure better performance of the model.
4 Experimental Validation
The performance of the proposed method is shown by the set of tests according to the ability to correctly diagnose the provided skin tissues. These results present a high classification accuracy, which proves the robustness of the used artificial neural network architecture.
Unsupervised Method Based on Superpixel Segmentation for Corpus
However, in many sagittal brain MRI slices, the fornix appears in the vicinity of the CC with a similar intensity (Fig.1) [2]. This study demonstrated that the total volume of the CC and its sub-regions is correlated with autism severity [8].
3 Proposed Method
CC Segmentation of the Midsagittal Slice
CC Parcellation
It is an adjustment of K-tools for generating superpixels to be faster than existing methods, more memory efficient while significantly improving segmentation accuracy. Second, a combination of color and spatial proximity is achieved by a weighted distance measure that allows both control over superpixel size and compactness.
4 Experimental Results
Qualitative Evaluation
In fact, according to our collaborating clinician expert, the CC shape and subdivision are well defined and the delineated CC area properly represents the five anatomical subdivisions of the CC, especially the critical ones: the rostrum and splenium. The fornix has been correctly removed from the CC region and the obtained CC parcellation shows an accurate subdivision of CC into five regions within MRI scans of the brain, without penetrating the irrelevant neighboring structures.
Quantitative Evaluation
Extensive experiments and quantitative comparisons with corresponding CC parcellation methods proved the accuracy of the proposed method on two challenging benchmark datasets. Witelson, S.: Hand and gender differences in the isthmus and gender of the human corpus callosum: a postmortem morphological study.
Using Learning Techniques to Observe Elderly’s Behavior Changes over Time
In this article, we focus on the problem of learning smart home sensor data describing the activities of the elderly. Our aim in this work is to propose an approach to identify periods when behavioral changes occur and detect anomalies in this period (e.g. older people sleep less and less every month).
2 Related Works
This daily score variation provides a global vision of the behavior of the elderly person over a period of time. Our goal in this work is not only to detect sudden changes, but also to analyze the possible evolution of the behavior over a long period of time.
3 Our Approach
Activities and Daily Behavior Pattern
Working in [2,3,5–8] consider home sensors to monitor daily activities, but do not analyze all activities of the older person at the same time. The user can perform the same activity at different times (for example, going to the toilet), but some activities only occur at specific times of the day (for example, having breakfast).
Normal Behavior Pattern for the Elderly
The start time and duration of each activity instance can be recorded by the user, or better detected by an activity recognition system based on sensors in the home. Using the same training data to model regularities in the three activities studied, we calculate the frequency, per n hour, of the "going to the toilet" activity.
Elderly’s Behavior Change Detection
It is the percentage of the duration of the activity in the observed day compared to the duration of the same activity in the usual behavior pattern. The duration estimate (80.95% in our case) is the percentage of the duration of the "eating lunch" activity in the observed day compared to the duration of the same activity in the usual behavior pattern.
4 Use Case
Dataset
Within home activities, anomalies can be classified as point, collective and contextual anomalies. For the activity “going to the toilet”, which occurs several times a day, we model and compare frequencies between a routine day and the observed day of the behavior change period.
Learning for Building the Normal Behavior Pattern
DBSCAN has two parameters; one is min pt which is the minimum number of points in a cluster and the other is Eps which is the maximum distance between two data points for them to be considered in the same cluster. We then eliminate point anomalies and calculate, for each activity in the training data set representing one month of collected data, the mean start time and mean duration.
Behavior Change Period and Anomalous Activities
The rows of the matrix contain types of outcomes (formation, change or reinforcement) and the columns contain types of change (observance, behavior or attitude). The development of MV is adaptable to the input spatial data in closed spaces and the profile of the inhabitant.
A Novel On-Wrist Fall Detection System Using Supervised Dictionary Learning
Technique
In the same direction, in this article we propose a new wrist-based fall detection system based on Supervised Dictionary Learning (SDL) to autonomously generate an optimal selection of features that best represent the acquired data. Indeed, the work presented here is an extension of previous research [19] that implemented a motion decomposition method to extract features (directional components and body orientation) and machine learning algorithms for fall detection based on a wearable device on the wrist.
2 Theoretical Background and Related Work
Wearable Fall Detection System
Dictionary Learning for Classification
The latter use label information in learning dictionary atoms, sparse vector coefficients, or both. Assuming that SDL methods and sparse representation differ in the way they use class labels, we will detail three of the most popular SDL algorithms, namely SRC, FDDL, and LRSDL.
Low-Rank Shared Dictionary (LRSDL)
Thus, the learned dictionary D = [D1,D2, .,Dm], where Di is a subdictionary corresponding to the classi, powerfully represents interclass similarity and intraclass variance. To describe FDDL more formally, assume X= [X1,X2, .,Xc], as the training examples are grouped according to the classes they belong to and c is the total number of classes.
3 Proposed Dictionary Learning Method
- Dataset
- Data Preprocessing
- Dictionary Learning for Fall Detection
- Performance Metrics
- Experimental Configuration
- Experimental Result
We will only acknowledge the vertical component of the motion and orientation decomposition, as it reached the best results in the latter work. In the training phase, the goal of the SDL algorithm is to map the low-dimensional training.
