Advances in Computer Vision and Computational Biology

There are more than 60 subtypes of Lymphoma. This diversity usually requires a specialised pathologist for diagnosis. We aimed to investigate the effectiveness of Artificial Neural Networks (ANNs) and Deep Learning at Lymphoma classification. We also sought to determine whether Evolutionary Algorithms (EAs) could optimise accuracy. Tensorflow and Keras were used for network construction, and we developed a novel framework to evolve their weights. The best network was a Convolutional Neural Network (CNN); its tenfold cross-validation test accuracy after training and weight evolution was 95.64%. The best single run test accuracy was 98.41%. This suggests that ANNs can classify Lymphoma biopsies at a test accuracy higher than the average human pathologist. The EA consistently improved accuracy, demonstrating that they are a useful technique to optimise ANNs for Lymphoma classification.

We propose a novel particle filter for convolutional-correlation visual trackers. Our method uses correlation response maps to estimate likelihood distributions and employs these likelihoods as proposal densities to sample particles. Likelihood distributions are more reliable than proposal densities based on target transition distributions because correlation response maps provide additional information regarding the target’s location. Additionally, our particle filter searches for multiple modes in the likelihood distribution, which improves performance in target occlusion scenarios while decreasing computational costs by more efficiently sampling particles. In other challenging scenarios such as those involving motion blur, where only one mode is present but a larger search area may be necessary, our particle filter allows for the variance of the likelihood distribution to increase. We tested our algorithm on the Visual Tracker Benchmark v1.1 (OTB100) and our experimental results demonstrate that our framework outperforms the state-of-the-art methods.

For recognizing speakers in video streams, significant research studies have been made to obtain a rich machine learning model by extracting high-level speaker’s features such as facial expression, emotion, and gender. However, generating such a model is not feasible by using only single modality feature extractors that exploit either audio signals or image frames, extracted from video streams. In this paper, we address this problem from a different perspective and propose an unprecedented multimodality data fusion framework called DeepMSRF, Deep Multimodal Speaker Recognition with Feature selection. We execute DeepMSRF by feeding features of the two modalities, namely speakers’ audios and face images. DeepMSRF uses a two-stream VGGNET to train on both modalities to reach a comprehensive model capable of accurately recognizing the speaker’s identity. We apply DeepMSRF on a subset of VoxCeleb2 dataset with its metadata merged with VGGFace2 dataset. The goal of DeepMSRF is to identify the gender of the speaker first, and further to recognize his or her name for any given video stream. The experimental results illustrate that DeepMSRF outperforms single modality speaker recognition methods with at least 3% accuracy.

Image watermarking based on deep learning has been proposed in the last few years. Typical framework for image watermarking consists of embedding network, extracting network, and attack stimulating module. Adversarial discriminator is sometimes used to make watermarked images much more similar to cover images. To improve the robustness of CNN-based work against attacks of different type and strength, we proposed a novel model, introducing recover module into the framework to compensate some damaged watermark information during attacks and improve extracting accuracy. For non-differentiable JPEG compression, we propose a new approximation approach based on previous methods. Experimental results show that the proposed model performs better than the state of the art in bit accuracy of message extraction while the image quality does not deteriorate.

In today’s world, plant diseases are a major threat to agriculture crops and their production rate. These are difficult to spot in early stages and it’s not feasible to inspect every leaf manually. We tested different convolutional neural networks on their ability to classify plant diseases. The best model reaches an accuracy of 99.70%, made with a deep training method. We also developed a hybrid training method, reaching a 98.70% accuracy with faster training times, reducing the gap between accuracy and training time. This was made possible due to the freezing of layers at a predefined step. In general, detecting plant diseases using deep learning models is an excellent approach and much more practical than detection with the human eye.

The visual quality of stitched images plays an important role to provide the high-quality immersive viewing experience of Virtual Reality (VR) contents. There are several image stitching algorithms that generate panoramas from multiple images taken with different cameras and angle of views. The performance of these stitching algorithms can be measured by estimating the quality of generated panoramas. This paper presents a segmentation-based Stitched Image Quality Assessment (SIQA) approach that captures the blended distortion in stitched images and segments the distorted region using binary mask. The segmented regions provide the location and total area of the distorted region. The results obtained from the experimental evaluation validate the reliability of our method for capturing the blended distortions in stitched images.

As high intraocular pressure (IOP) is the main cause of glaucoma which can result in irreversible vision loss, early detection is extremely important for prevention. This paper is a research work in progress, built upon our prior work, to distinguish healthy eye images from high IOP cases using a deep learning approach solely from frontal eye images. We propose a novel computer vision-based technique using a convolutional neural network (CNN) to extract common features of high IOP and glaucoma cases automatically from frontal eye images. The dataset used in this work contains 473 normal and high IOP frontal eye images. The proposed idea has the potential to minimize the patient’s presence at healthcare facilities and prevent glaucoma causes by automating early detection.

Apple leaf disease is one of the main factors to constrain the apple production and quality. It takes a long time to detect the diseases by using the traditional diagnostic approach; thus, farmers often miss the best time to prevent and treat the diseases. The classification of apple leaf disease is an essential research topic. Compared with the conventional approaches depending upon either region segmentation or end-to-end learning of full image by a neural network, we propose superpixel-based disease classification. CNN (convolutional neural network) is used as a classifier. Based on PlantVillage images, we show that the proposed method is able to match, achieving a 92.43% accuracy as well as F1 score of 0.93 compared with 98.28% and 0.98 of full image.

The pursuit in similar image detection is constantly increasing in the computer vision field. This is even more prominent in the medical imaging field. Medical image sets such as cancerous pigmented skin lesions have a long tendency of including images with multiple modalities. The diagnostic algorithms such as classifiers that categorize the skin lesions rely on unique set of images for best performance. In this paper we developed a distance-based approach to image similarity detection. We applied six methods of image distances which utilized features from image histogram. The sensitivity and effectiveness of each method as well as the determination of threshold is discussed.

In recent times, object detection and pose estimation have gained significant attention in the context of robotic vision applications. Both the identification of objects of interest as well as the estimation of their pose remain important capabilities in order for robots to provide effective assistance for numerous robotic applications ranging from household tasks to industrial manipulation. This problem is particularly challenging because of the heterogeneity of objects having different and potentially complex shapes, and the difficulties arising due to background clutter and partial occlusions between objects. As the main contribution of this work, we propose a system that performs real-time object detection and pose estimation, for the purpose of dynamic robot grasping. The robot has been pre-trained to perform a small set of canonical grasps from a few fixed poses for each object. When presented with an unknown object in an arbitrary pose, the proposed approach allows the robot to detect the object identity and its actual pose, and then adapt a canonical grasp in order to be used with the new pose. For training, the system defines a canonical grasp by capturing the relative pose of an object with respect to the gripper attached to the robot’s wrist. During testing, once a new pose is detected, a canonical grasp for the object is identified and then dynamically adapted by adjusting the robot arm’s joint angles, so that the gripper can grasp the object in its new pose. We conducted experiments using a humanoid PR2 robot and showed that the proposed framework can detect well-textured objects, and provide accurate pose estimation in the presence of tolerable amounts of out-of-plane rotation. The performance is also illustrated by the robot successfully grasping objects from a wide range of arbitrary poses.

Symmetry is a very common geometric feature in natural objects and, highly marked, in artificial objects; particularly, mirror symmetry is a relevant topic in fields such as computer vision and pattern recognition. Previous work on symmetry has shown that there are useful solutions to the problem of mirror symmetry detection; however, there are still important challenges to successfully model symmetry objects and properly detect multiple axes. And there is also the challenge of assigning a level of symmetry to quasi-symmetrical objects. In this paper, we propose an algorithm that detects level and axes of symmetry for quasi-mirror and mirror symmetry in 2D contours represented by AF8 code.

In this study, a superpixel-based stereoscopic video saliency detection approach is proposed. Based on the input stereoscopic video sequences containing left-view and right-view video sequences, a sequence of right-to-left disparity maps is obtained. First, the simple linear iterative clustering (SLIC) algorithm (Achanta, IEEE Trans Pattern Anal Mach Intell, 34(11):2274–2282, 2012) is used to perform superpixel segmentation on all video frames. Second, the spatial, temporal, depth, object, and spatiotemporal features are extracted from video frames to generate the corresponding feature maps. Third, all feature maps are concatenated and support vector regression (SVR) learning using LIBLINEAR tools is employed to generate the initial saliency maps of video frames. Finally, the initial saliency maps are refined by using the center bias map, the significant increased map, visual sensitivity, and Gaussian filtering. Based on the experimental results obtained in this study, the performance of the proposed approach is better than those of three comparison approaches.

Pollution encompasses any substance that has a negative impact on the environment or the organisms that live within an affected area. Marine environmental pollution includes debris, which can be natural or man-made. In this paper, four distinct debris scenes are analyzed using image processing tools to detect man-made marine debris with different image and object properties. The scenes vary in their challenges from multiple floating surface and subsurface debris to single pieces of debris. A successful image processing chain is described that includes image preprocessing such as conversion to HSV color model, filtering, thresholding, and post-processing operations such as blob analysis and statistical computations to detect and characterize the man-made debris in each scene. We demonstrate detection of multiple debris and computation of its percent cover of the scene, its size distribution, as well as identification of single debris pieces. The applied methods demonstrate successful debris detection and characterization in each scene, which can be extended to other debris images for automated detection.

A new method for polyhedral approximation is presented in this paper. The representation of 3D objects is a complicated task. This is the reason why the object is organized as a slices set. The proposed method takes the slices, obtains the chain code from the contour, and uses the existing context-free grammar method to find the dominant points from contour of each slice, obtaining a point cloud from the selected slices. These dominant points are strategically joined to create the approximate polyhedron of the object. Finally, we adapt an existing error criterion to evaluate the approximated polyhedron with the original object.

Real-time action recognition on smartphones and wearable cameras is a challenging task. The effect of camera motion on action recognition algorithms is not trivial. To our advantage, smartphones and wearable devices are often augmented with depth, audio, geo-location, and inertial data, which can be incorporated into action recognition framework. Our study aims to quantify the effect of ego-centric motion on standard action recognition algorithms. As a part of the study, we collected multi-sensor video dataset with seven actions from the same subjects captured in three different camera motion scenarios—minimal, pure rotational and both rotational and translational. We plan to create a multi-sensor fusion based action recognition framework to improve the recognition accuracy on smartphones and wearable cameras, which are equipped with inertial sensors. We present preliminary experiment results on our multi-sensor video dataset.

Aerial surveillance systems have become an important aspect in a range of civilian and military applications. For instance, reliable location data of individuals or objects is essential for use cases such as traffic surveillance, search and rescue, relief efforts after natural disasters, or patrol and border control missions. Unmanned aerial vehicles (UAVs) provide great value in such situations but conventionally require at least one human operator whose insights and intuition are necessary for performing the task. With this contribution, we conceptualize a functional design that leverages performance models to track multiple targets and generate situational awareness autonomously. Further, we devise a testing harness for different scheduling schemes and provide some preliminary experimental data on basic scheduling policies found in literature.

Due to the finite depth of field of optical lenses, it is difficult to make all objects in an image sharp and clear. Only objects within the depth of field are in focus and sharp, while the others are defocused and blurred. In this study, a superpixel-based multi-focus image fusion approach is proposed. First, each multi-focus source image is performed superpixel segmentation, and the saliency, depth, and difference image information are computed. Second, each superpixel is classified into one of three types (focus, defocus, and undefined), and each undefined superpixel is determined as focus or defocus by sum-modified-Laplacian (SML). The initial focus maps are estimated and refined by matting Laplacian-based image interpolation. Third, the boundaries between focus and defocus regions are employed to generate the weighting maps, followed by fused image generation. Based on the experimental results obtained in this study, the performance of the proposed approach is better than those of five comparison approaches.

This study was conducted to demonstrate the potential for magnetic fields to serve as a method of remotely sensing electronic activity and to evaluate the potential for classifying the electronic activity. To demonstrate this, a radio frequency generator, antenna, and oscilloscope were placed inside a Faraday cage with varying frequencies transmitted in the range of 1–1000 Hz. A standard radio frequency antenna and magnetic loop antennas were placed outside the Faraday cage and the results were compared to each other as well as natural ambient signals. The results showed positive detection of magnetic field activity outside of the Faraday cage in the transmitted frequencies, with no detection with the radio frequency antenna. They also demonstrated the inability of a Faraday cage to attenuate magnetic fields of objects inside the cage. Errors that produced anomalies in the first attempt served to further validate the collection of the data by generating positive detection on both antennas. Ultimately the use of magnetic field antennas to detect electronic activity demonstrated potential use in a radio frequency adverse environment.

In this chapter, we propose an optimization of the a-contrario clustering method using the probabilistic Guillaume Khenchaff Measure (MGK) quality technique. A-contrario is used for tracking salient objects in the scene in real time. This method analyzes the data contained in a motion vector, which contains the scattered optical flow accumulated points of interest. The aim of our study is to improve the first results obtained from the Number of False Alarm (NFA) criterion by using MGK to bring together the group of points endowed with a coherent movement of the binary tree. The idea is to isolate dynamic points so that we can use static points in the future.

Vehicle mirrors can reflect objects behind the cars and play an important role in driving security. In manufacturing process of vehicle mirrors, certain tasks operated unusually will cause producing surface and profile defects on vehicle mirrors. Those appearance defects sometimes will severely have an impact on standard of the mirror reflection and grow the driving hazard. At traditional examination of vehicle mirrors in manufacturing process, almost all works are performed by human examiners. This study works toward investigating the automatic appearance defect detection of vehicle mirrors. We propose a defect enhancement technique based on Fourier high-pass filtering and the convex hull arithmetic to inspect appearance defects on vehicle mirrors. This approach only utilizes their own information of testing images to judge whether there are any irregular appearance changes without the need of standard patterns for matching. Experimental results show that performance of the Fourier-based approach in the defect detection is effective and efficient.

The metrics or measures used in evaluating image quality usually serve applications that aim at enhancing image appearance or at preventing the image quality from degrading after image processing operations. On the other hand, the main goal of cryptography applications is to produce unrecognizable encrypted images that can resist various kinds of attacks. Furthermore, cryptography must consider extra measures such as keyspace and key sensitivity. This paper discusses the most useful quality metrics used in image cryptography and explains the type of values that indicate good encryption according to each metric. These metrics include statistical analysis measures, sensitivity measures, and other metrics.

In this paper, a study on the use of ICT support to help therapists and Systemic Sclerosis (SSc) patients in the recovery phase is described. The ReMoVES platform is conceived in the field of assistive computing technologies and delivers engaging exergames to be performed in conjunction with traditional rehabilitation both at home and at clinical centers, thus enabling the continuity of care. The present work refers to the implementation and integration in the ReMoVES platform of standard hand rehabilitative exercises for SSc that typically involves repetitive movements. Data related to game sessions are collected and analyzed for assessing the patients’ conditions.

The development of an objective algorithm to assess craniosynostosis has the potential to facilitate early diagnosis, especially for care providers with limited craniofacial expertise. In this study, we process multiview 2D images of infants with craniosynostosis and healthy controls by computer-based classifiers to identify disease. We develop two multiview image-based classifiers, first based on traditional machine learning (ML) with feature extraction, and the other one based on CNNs. The ML model performs slightly better (accuracy 91.7%) than the CNN model (accuracy 90.6%), likely due to the availability of a small image dataset for model training and superiority of the ML features in differentiation of craniosynostosis subtypes.

This paper addresses the problem of identifying two main types of lesions—Exudates and Microaneurysms—caused by Diabetic Retinopathy (DR) in the eyes of diabetic patients. We make use of Convolutional Neural Networks (CNNs) and Transfer Learning to locate and generate high-quality segmentation mask for each instance of the lesion that can be found in the patients’ fundus images. We create our normalized database out of e-ophtha EX and e-ophtha MA and tweak Mask R-CNN to detect small lesions. Moreover, we employ data augmentation and the pre-trained weights of ResNet101 to compensate for our small dataset. Our model achieves promising test mAP of 0.45, altogether showing that it can aid clinicians and ophthalmologist in the process of detecting and treating the infamous DR.

Hip fracture incidence increases with age and is common among older population. It causes significant problems as there is an increased risk of mortality, restriction of the movement and well-being of the injured people, loss of independence, and other adverse health-related concerns. Rehabilitation plays a significant role in recovery and improving the physical functionality of patients in all stages of care. This is through supporting early postoperative mobilization to secondary prevention using balance and exercise activity movements. Many studies have analyzed the postoperative effect of physiotherapeutic exercise and mobility during hip fracture rehabilitation process. Nevertheless, none of them has highlighted the key stages and activities progression pathway involved during postoperative rehabilitation recovery process. Therefore, it is not clear which care and rehabilitation services accomplish the suitable outcomes for the people undergoing rehabilitation. This article presents conceptual model for the process of the postoperative hip fracture rehabilitation. The model reflects the key stages a patient undergoes straight after hospitalization. The concept could support the development of an online activity monitoring system that could track the vital changes or ongoing improvements taking place in a patient’s well-being as they go through the rehabilitation program. This should also offer vital information to clinicians, related hospitals people, caregivers, as well as the patient’s him/herself. Another area of interest here is that of unfolding the progressive improvement in related muscles in reference to the stages of the program and how this can evolve to offer more precise interaction with the rehabilitation process.

As human beings live longer, the number of people diagnosed with dementia is growing. Many studies have proved that dementia tends to degenerate cognitive abilities. Since dementia patients endure different types of symptoms, it is important to monitor dementia patients individually. Furthermore, senior citizens are generally lack of understanding technology, which brings a low self-motivation to use technologies. To enhance the cognitive abilities of senior citizens, we propose a mobile platform called ReSmart which embeds six distinct levels of the brain training task, based on five cognitive areas to detect different types of individual symptoms. Those brain training tasks are presented in a game-like format that aims to not lose the elders’ motivation for technology use and keeping interest.

ActiviX is an at-home mental health support tool that aids people who suffer from mental health by encouraging them to perform small but meaningful tasks in hope of leading them to a happier life. ActiviX focuses on the self-care and productivity of users by keeping track of what important tasks they complete throughout their day. These tasks are defined by the user and can range from maintaining good hygiene to doing homework or eating three meals a day and staying hydrated. If the system detects that the user is skipping out on these tasks, ActiviX reminds the user to complete them. Additionally, if it detects that the user is doing well and their mood is improving, it will give bits of motivation to encourage them to keep up the good work.

While predicting the evolution of Covid-19 infections on a medium and long term is mostly related to epidemiological models and considerations from disciplines of virology, immunity, and epidemiology, the short-term (few days ahead) evolution of Covid-19 curves can help draw the trend and understand if an outbreak is still in a dangerous exponential evolution or not. In this chapter, we show how we transformed the data and applied polynomial curve fitting to help on this objective, how we discovered the most appropriate polynomial, and the visualizations we got from all the alternatives.

Lack of physical activity (PA) is one major contributor to the high prevalence of modern-day chronic diseases. In this study, we design and implement an interactive PA motivation program for older adults (65+) and people with chronic conditions by adapting a commercially available artificial intelligence voice-assisted (AIVA) application and activity tracker (Alexa/Echo Show/Fitbit). Customized features are founded on persuasive technology theory and techniques- and evidence-based PA programs. A pilot study with four older adults for 5 days demonstrated moderate to good usability, learnability, satisfaction, and performance and key-persuasive techniques that enhance user experience and goal achievement. The study provides a tested model for PA behavior change using AIVA for feedback, education, and motivational guidance.

The use of big data analytics tools and Machine Learning techniques in identifying and predicting opioid use disorder is a relatively new and emerging area. Previous studies analyzing trends in the opioid crisis have identified an increasingly expensive and worsening epidemic. Many factors contribute to opioid use, abuse, and addiction. Opioid addiction is a complex disease involving various physiological pathways as well as environmental factors. There is some evidence to suggest that people with low education levels and high unemployment and poverty levels are at higher risk of opioid abuse. In this paper, we evaluated different conventional Machine Learning models including Support Vector Machines (SVM), Decision Tree, and Logistic Regression and advanced algorithms like Gradient Boosting. The models were developed to predict opioid abuse disorder using county-level education, poverty, and unemployment data. In contrast, the results suggest that populations with higher socioeconomic status are at greater risk of opioid abuse disorder compared to individuals with lowers. This can be attributed to underlying factors not previously captured increased availability of opioids and resources to acquire them. Identifying which areas and populations are at higher risk of opioid abuse disorder and underlying contributing factors will help inform judicious effective policies, programs, and solutions to tackle the worsening health crisis.

Mobile applications for healthcare must be both secure and usable while operating within the boundaries of a healthcare organizational network to access critical information resources, such as the electronic health record (EHR). However, in an increasingly interorganizational mobile work environment, organizations are seeking provider-centered apps for continuity of care between and across organizations. Emergency medical services (EMS) provide one example where a single transport agency may direct patients to any number of separate health systems within a region. This research reports on the design and live field testing of an interorganizational mobile app for EMS. Participants included 20 ambulances and seven hospital emergency departments transmitting 1513 live patient records with multimedia data. Evaluation results were used to present a heuristic for achieving usability and security goals for practitioner-oriented interorganizational mobile health applications, especially for time- and information-critical contexts.

The automatic speech recognition has achieved quite good performance in the medical domain in the past several years. However, it is still lacking of enough practical solutions with considering the characteristics of real applications. In this work, we develop an approach to automatically generate semantic-coherent ultrasound reports with voice input. The solution includes key algorithms based on a proposed semantic tree structure. The radiologists do not need to follow the fixed templates. They just need to speak their specific observations for individual patients. We have carried out a set of experiments against a real world thyroid ultrasound dataset with more than 40,000 reports from a reputable hospital in Shanghai, China. The experimental results show that our proposed solution can generate concise and accurate reports.

In this chapter, we describe the application of the technologies of the Internet of things (IoT) to the management of hospital beds. Specifically, it seeks to monitor the status of hospital patients assigned by their doctors to bed rest based on their medical conditions on sensor data collected by embedded pressure sensors and provide real-time alerts to the medical staff. The potential for injuries from prescribed bed-rest patients vacating the hospital bed and falling is very serious. The injuries often result in additional complications to the underlying health condition requiring the prescribed bed rest. The proposed IoT bed-rest management system will alert the medical staff immediately when a bed-rest patient vacates the bed and allow them to take immediate remedial action. The research consists of two parts. The first part created IoT-connected pressure sensors and devices used to capture the patients’ presence or absence on the bed and send the data to a central server. The second part developed the medical staff alert application that runs on mobile phones and consoles located in the nurses’ stations, that receive the information from the server. This system can also monitor the movement of comatose patients, who need to be moved frequently by the medical staff to avoid pressure ulcers. The server can automatically inform the medical staff when a patient is scheduled to be moved and alert the staff when a scheduled movement is overdue for a set time.

In this study, we develop a predictive model for the length of stay (LOS) for chronic obstructive pulmonary disease (COPD) patients using administrative, clinical, and operational data from a large teaching hospital in the southeastern United States. To address the issue of a large percentage of missing values for several predictor variables in the clinical data set, we carry out multiple imputation of the missing values with the bootstrapping and expectation–maximization (EM) algorithm from Amelia package in R and perform variable selection with the Boruta algorithm. We employ generalized linear models (GLM) and random forests to model the response variable and perform model comparison based on their generalization errors.

In this paper we study the use smart glasses in classifying simulated epileptic seizure signals. We train a patient specific classifier using features extracted from an inertial measurement unit signals. For performance evaluation, we use the accuracy as well as the loss function values and Root-Mean-Square Error (RMSE). Long short-term memory (LSTM) neural network is used on the data collected from the smart glasses. Orhan et al. (Exp Syst Appl 38:13475–13481, 2011), Samiee et al. (IEEE Trans Biomed Eng 62:541–552, 2015)

This paper describes the use of pre-trained model classifiers in detecting epileptic seizures. A patient specific classifier was trained using features extracted from the intracranial electroencephalogram (iEEG) signals. Both accuracy and the loss function are used for performance evaluation.In order to use the pre-trained models that are commonly used for image classification the time series iEEG signal was converted into a spectrogram image associated with that signal. The model was then trained and our results are shown below.

This chapter constitutes a first step toward a wearable epileptic seizure prediction device. Since recording electrocardiogram can be accomplished fairly easily, we look into the existing correlation between epileptic pre-ictal states and heart rate variability. The intervals of extreme noise may corrupt the electrocardiogram data during the seizures, and this means that we are able to use a machine learning and specifically deep learning techniques to detect the pre-ictal aka pre-seizure states. The experimental results show particularly good results in terms of prediction performance. They also show the importance of a specific training for each patient. In this chapter, we analyzed the cardiac dynamics in patients with partial epilepsy. By doing so, we discovered transient but prominent low-frequency heart rate oscillations immediately following seizures in some patients. These features have been used for understanding cardiac and neuro-autonomic instability in epilepsy and also for classifications of such heart rates.

People with Parkinson’s disease must be regularly monitored by their physician to observe how the disease is progressing and potentially adjust treatment plans to mitigate the symptoms. Monitoring the progression of the disease through a voice recording captured by the patient at their own home can make the process faster and less stressful. Using a dataset of voice recordings of 42 people with early-stage Parkinson’s disease over a time span of 6 months, we applied multiple machine learning techniques to find a correlation between the voice recording and the patient’s motor UPDRS score. We approached this problem using a multitude of both regression and classification techniques. Much of this paper is dedicated to mapping the voice data to motor UPDRS scores using regression techniques in order to obtain a more precise value for unknown instances. Through this comparative study of variant machine learning methods, we realized some old machine learning methods like trees outperform cutting edge deep learning models on numerous tabular datasets.

Environmental change has become an important issue in many African countries with much recent discussion focusing on the harmful health and environmental impact resulting from human activities such as fossil fuel, oil spillages and others in the Global South. Many efforts are being carried out to deal with the issue of climate change and environmental degradation. One effort is the use of information and communications technologies (ICTs) as an instrument for environmental protection and the sustainable use of natural resources. ICT can be leveraged to tackle harmful environmental change. This chapter offers an overview of how ICTs can be used to benefit the environment in Nigeria. Guidance on how to leverage technology for the good of the environment in Nigeria is provided.

Primate models are important for understanding human conditions, especially in studies of aging, pathology, adaptation, and evolution. However, how to integrate data from multiple disciplines and render them compatible with each other for datamining and in-depth study is always challenging. In a long-term project, we have started a collaborative research endeavor to examine the health history of a free-ranging rhesus macaque colony at Cayo Santiago and build a knowledge model for anthropological and biomedical/translational studies of the effects of environment and genetics on bone development, aging, and pathologies. This chapter discusses the conceptual design as well as the prototyping of this model and related graphical user interfaces and how these will help future scientific queries and studies.

Early detection of breast cancer is recognized as the best way to reduce mortality. Mammography is widely used to detect cancer. This chapter discusses our underlying efforts to implement a mammography data warehouse to encourage medical and investigative activities. The data warehouse is a reliable combination of information from many sources. We are planning an infra-basic information system by merging different types of breast imaging information, a decent variety of existing health systems, into an advanced data warehouse center. Different types of breast imaging containing demographic data, family history, mammograms, and textual reports will be obtained from Salah Azaiz Cancer Institute picture archiving and communication system (PACS) modules, as well as the radiological information system (RIS). This research paper proposes a data warehouse analytical framework for exploring and analyzing data related with breast cancer in Tunisia for decision support.

Nowadays, the emergence of smart technologies in healthcare domain is revolutionizing all aspects of patients’ daily life. Indeed, the continual increase of patients with chronic diseases has revealed two major challenges: improving the patients’ living quality, which has been motivated by their growing need to be cared in a family environment, and reducing the costs of care. Remote patient monitoring (RPM) at home represents a promising opportunity to face these challenges. It is mainly based on using smart devices such as sensors in order to monitor the patient’s status anywhere and at any time and to detect earlier any critical health situation to trigger different actions accordingly. Based on this context, we are designing a system capable of offering services in order to monitor and assist patients at home. Indeed, these services could actuate different actions according to detected situations. But it is necessary to notice that all patients do not have the same needs and preferences. So the system should be able to cover all characteristics that differentiate each patient as well as the devices that are used in their environment.

Accessing a patient’s information across data sharing networks is a challenging task. For a client application to request a patient’s data, it has to first refer to a centralized repository for demographic information to identify the patient. Then the search will be continued for patients’ clinical and medical data that may exist in different centralized data sharing networks. This approach poses a risk on data availability especially in emergency instances because centralized data sources can be attractive targets for cyberattack (Peterson et al., J. Med. Syst. 63:425–432, 2016) or can be a single point of failure. Other problems can be data privacy and security associated with the centralized authority and governance of data (Wu and LaRue, Int. J. Nurs. Sci. 4:410–417, 2017). In this chapter, we introduce a cross-blockchain-based data search service that avoids centralized data risks. This search service consists of emergency data services that enable first responders to request and receive relevant emergency data across multiple Hyperledger Fabric (HLF) networks. It also allows first responders’ care reports to be sent back to the networks where patient’s data is retrieved from. Patients’ treatment data is recorded and updated on the ledger. We have implemented our approach by creating two HLF networks consisting of two hospitals and one client application, which enables first responders to look up a patient’s data and add the care report by connecting to these networks.

Robotic process automation (RPA) is a specialized software robot widely used for facilitating repeated tasks. The RPA can augment medical staff in data gathering, analysis, and reporting process. This chapter presents an RPA framework for a mobile glaucoma screening system. With the RPA, fundus images and clinical history of patients are gathered and submitted to the machine learning-based glaucoma diagnosis. If the preliminary diagnosis results in a severe condition, the application directly prompts the eye specialists for timely verification and treatments. The application facilitates the specialists in adjusting clinical parameters, making final diagnosis decision remotely, gathering and reporting information and decision, and scheduling the next hospital visit or re-imaging. The integration of the RPA framework within the eye screening system significantly reduces costs and time of operations, allows timely treatments, improves customer experiences, and promotes feasibility in large-scale population glaucoma screening.

There is an enormous number of healthcare analytics applications in existence which have been embedded into healthcare systems with varying degrees of success. One of the key challenges is their need for access to sensitive patient data in a healthcare system that has a multitude of healthcare applications. This chapter introduces a new theoretical framework as an architecture in which Healthcare 4.0 applications can operate. The framework proposes using Apache Kafka as the core technology for creating data integration pipelines with the goal being to bring standardisation into the healthcare systems. The architecture offers a safe and secure environment in which multiple applications and algorithms from different organisations can seamlessly co-exist.

Prostate cancer screening and diagnosis remain controversial due to the debate regarding overdiagnosis and subsequent overtreatment of prostate cancer. Reducing unnecessary prostate biopsies is a crucial step toward reducing overdiagnosis. As we move toward more personalized medicine and individualized medical decision-making, there is a fundamental need for better risk assessment tools that can aid patients and physicians in this decision-making process. The presented work here seeks to construct risk prediction models to predict the presence of prostate cancer, clinically significant cancer (Gleason score ≥7), and unfavorable pathology (pT3a or pT3b and Gleason grade group ≥3) on initial biopsy. Such multivariable risk prediction models can be used to further aid in patient counselling for those undergoing prostate biopsy.

In this work, a natural language processing algorithm is proposed for the analysis of outcomes of a member engagement specialist’s cold calls to prospective mental and behavioral care patients. The purpose of the call is to introduce prospective members to specially designed mental healthcare therapy and get them to enroll in the program. The proposed approach is based on keyword analysis, and the results are provided on the analysis of 9254 transcribed voice calls using Google Cloud Platform (GCP) of varying duration between 30 and 50 minutes. Additionally, qualitative design was used to identify keywords and phrases for high-performer and low/mid-performer member engagement specialists.

Smart-health has the potential to create a unique platform for monitoring health. Personalized healthcare using smartphones offer seamless solutions, especially for circumstances such as the COVID-19 pandemic where physical distancing is inevitable. This paper investigates the efficiency of general smartphone-based healthcare monitoring applications (apps) from a system dynamics perspective. The ongoing research effort introduces a causal model to investigate the factors and inter-relationships that impact the efficiency of smartphone-based healthcare monitoring apps. A careful study suggests that the most important factors of such systems include patient well-being, satisfaction, cost, and performance measure factors (i.e. response speed, accuracy). The proposed idea provides a novel insight of the dynamics of the model to assess the efficiency of various smartphone-based healthcare monitoring apps.

Covid-19 pandemic caused by the SARS-CoV-2 has claimed numerous lives around the world. We developed a novel predictive model based on machine learning algorithms to predict the mortality risk of patients with COVID-19. In this study, we used documented data of 117,000 patients worldwide with laboratory-confirmed COVID-19. This study proposes a predictive model to help hospitals and medical facilities decide who has higher priority to be hospitalized and triage patients when the system is overwhelmed by overcrowding. The results demonstrate 93% overall accuracy in predicting the mortality rate. We used a number of machine learning algorithms including artificial neural networks, support vector machine (SVM), and random forest to predict the mortality rate in patients with COVID-19. In this study, the most alarming symptoms and features were also identified.

Dental informatics is a relatively new research field, but its publication has been steadily increasing in the past decade. Altmetric is a platform that collects information from different web sources and combines it together into a citation-based metric for researchers, publishers, and institutions. It tracks the attention that research outputs (such as scholarly articles and datasets) receive online. Altmetric Explorer is an easy-to-use web-based platform that enables users to browse and report on all attention data for a given scholarly output (including journal articles and dataset). Given the nature of dental informatics, research is quite practical. The purpose of this study is to perform an Altmetric analysis to systematically study research trends on dental informatics. The study identifies various aspects of research outputs on dental informatics (attention scores, a timeline of mentions, yearly publication total, top ten journals, top 15 affiliation of first authors, and top five mention categories) that would be of interest to both researchers and practitioners in the field. In conclusion, dental informatics needs more publicity. Although there is an increase in multiple aspects (such as social media mentions and publications) in dental informatics research, it is not enough to match other popular fields.

Quantitative structure activity/property relationship (QSAR/QSPR) analysis is a major approach for computer-aided drug design and has also been studied in bioinformatics. Recently, a novel method has been proposed for inverse QSAR/QSPR using both artificial neural networks (ANN) and mixed integer linear programming (MILP), where inverse QSAR/QSPR is to infer chemical structures from given chemical activities/properties. However, the framework has been applied only to the case of acyclic chemical compounds so far. In this paper, we significantly extend the framework and present a new method for inverse QSAR/QSPR of monocyclic chemical compounds. The results of computational experiments using such chemical properties as heat of atomization, heat of combustion, and octanol/water partition coefficient suggest that the proposed method is much more useful than the previous method.

Naturally occurring bacterial immune system can be engineered for use in the mammalian genome editing. To assist with the design of new editing systems, we developed and evaluated three data-driven predictors of DNA targets in the mouse and human genomes. Long Short Term Memory network models outperformed classifiers trained with Support Vector Machines and Random forest algorithms. The hold-out accuracy of the deep learning classifier reached 81.6% for the mouse genome and 82.5% for the human genome. We also demonstrated that classification accuracy improves when sequences surrounding a mammalian target site are incorporated into the input vector of the neural network, reaching an accuracy of 83% for both organisms.

Conserved noncoding elements (CNEs) are non-protein-coding genomic regions that exhibit an extreme degree of conservation. CNEs are mostly clustered around the genes and found to play important roles in regulating the transcription processes. Identification of CNEs is, therefore, important for studying their functional properties. Most of the existing CNE-finding methods are pairwise alignment based, and their scalability suffers when multiple sequences need to be processed simultaneously. We propose a new efficient alignment-free CNE-finding method, MinCNE. It finds CNEs among multiple sequences where k-mers derived from each sequence are clustered using minhash signatures and locality-sensitive hashing. The performance evaluation demonstrated MinCNE to be computationally efficient without compromising the accuracy. The MinCNE source codes are available at https://github.com/srbehera/MinCNE .

Machine learning and the use of neural networks have increased precipitously over the past few years primarily due to the ever-increasing accessibility to data and the growth of computation power. It has become increasingly easy to harness the power of machine learning for predictive tasks. Protein structure prediction is one area where neural networks are becoming increasingly popular and successful. Although very powerful, the use of ANNs requires the selection of the most appropriate input/output encoding, architecture, and class to produce optimal results. In this investigation, we have explored and evaluated the effect of several conventional and newly proposed input encodings and the selected optimal encodings, window sizes, and architectures. We considered 11 variations of input encoding, 11 alternative window sizes, and 7 different architectures. In total, we evaluated 2541 permutations in application to the training and testing of more than 10,000 protein structures over the course of 3 months. Our investigations concluded that one-hot encoding, the use of LSTMs, and window sizes of 9, 11, and 15 produce the optimal outcome. Through this optimization, we were able to improve the quality of protein structure prediction by predicting the ϕ dihedrals to within 14∘–16∘ and ψ dihedrals to within 23∘–25∘. This is a notable improvement compared to previous similar investigations.

The extraction of palm region of interest (ROI) is one of the most important tasks in palm recognition because the accuracy of ROI detection affects directly the recognition performance. In contactless environment, the free hand posture of users causes much more difficulties to determine the accurate ROI among a number of challenges. To the best of our knowledge, the disadvantage in most conventional methods is that the fingers are required to spread and the hand faces the camera. This restricts the flexibility of the system. Besides, there is a lack of modalities that work well in the case of large hand rotation. This study proposed new robust palm ROI extraction method to handle the above challenges. The experimental results on a public benchmark dataset as well as our self-collected dataset validate the high accuracy in extracting palm ROI from the proposed method.

The fundamental clinical properties of cancer chemotherapy are investigated and demonstrated by utilizing a system of clinically plausible deterministic non-linear differential equations which depicts the pathophysiology of malignant cancers. In this mathematical model, the cytokinetic properties of normal cells, cancer cells, and the pharmacokinetics of the chemotherapy drug are described, respectively, by biophysically measurable growth parameters, stoichiometric rate constants and Michaelis–Menten type reaction profiles. Computer simulations have been conducted to elucidate various hypothetic scenarios when the model is configured with different parametric values, including the therapeutic efficacy of the use of stealth liposomes in high dose chemotherapy.

Cancer is one of the main causes of death worldwide. We know that a significant percentage of cancers,; including breast cancer, can be cured by surgery, or chemotherapy; therefore, its detection at an early stage of the disease is essential.Raman spectroscopy is an optical technique capable of measuring vibrational modes of biomolecules, allowing their identification from the correct location of the Raman bands, one of the main challenges is the elimination of spectral noise composed of (a) fluorescence background and (b) high frequency noise. In this articlechapter, we demonstrate that, using ANFIS (Neuro Fuzzy Adaptive Inference System) in combination with moving averages filter on the MATLAB multicore platform, we can eliminate these disturbances and optimize response time in the preprocessing of large volumes of data in order to achieve a high precision in the classification and diagnosis of breast cancer.

Protein–peptide interactions, in which one partner is a globular protein and the other is a flexible linear peptide, are important for understanding cellular processes and regulatory pathways and are therefore targets for drug discovery. In this study, we combined rigid-body protein–protein docking software (MEGADOCK) and global flexible protein–peptide docking software (CABS-dock) to establish a re-ranking method with amino acid contact profiles using rigid-body sampling decoys. We demonstrate that the correct complex structure cannot be predicted (< 10 Å peptide RMSD) using the current version of CABS-dock alone. However, our newly proposed re-ranking method based on the amino acid contact profile using rigid-body search results (designated the decoy profile) demonstrated the possibility of improvement of predictions. Adoption of our proposed method along with continuous efforts for effective computational modeling of protein–peptide interactions can provide useful information to understand complex biological processes in molecular detail and modulate protein–protein interactions in disease treatment.

The structural conduct and preference of a protein are highly sensitive to the environment accommodating it. In this study, the solvation effect of the structure and dynamics of a domain in the C-terminal of RVFV L protein was explored by molecular dynamics using implicit and explicit water. The force field parameters of explicit waters were taken from the TIP3P, TIP4P, SPC/E, SPCE/Fw, and OPC water models, and those of the peptide were taken from the AMBER ff14SB force field. The behavior of the peptide was also investigated in an implicit solvent environment using the generalized Born (GB) model by setting the dielectric constant to match that of experimental measurements of water. Several properties including the interaction energy between the peptide and solvent molecules are calculated. Structural characterization and clustering of the atomic trajectories enable a better understanding of the structural and dynamical behavior of the RVFV domain along time.

Genes and gene products interact in an integrated and coordinated way to support functions of a living cell. In this research, we analyze these interactions in 17 different types of cancers, focusing on the interactions presented in pathway maps in Kyoto Encyclopedia of Genes and Genomes repository. We extracted the gene-to-gene interactions from the pathway maps and integrated them to form a large integrated graph. We then utilized different techniques and filtering criteria to extract and shed lights on the gene-gene interaction patterns. We conclude that the graph motifs we identified in cancer pathways provide insights for cancer biologists to connect dots and generate strong hypotheses so further biological investigations into cancer initiation, progression, and treatment can be conducted effectively.

We utilize the GWAS Catalog to gather a set of common SNPs with high magnitude odds ratio values for an increased risk of asthmatic phenotypes. After collecting a set of SNPs, we parse it down to a subset of ten representative variations. Using the odds ratio and known prevalence of asthma in the population, we then calculate each SNP’s individual contribution towards the lifetime genetic risk of developing asthma. For a hypothetical European patient, we estimate that the SNP from our set with the smallest odds ratio value sees an increase of lifetime risk by 8%, while the SNP with the largest odds ratio value causes the risk of developing asthma to more than double. For a more complete and accurate prediction of asthma risk, additional work must be done to incorporate the contribution from less common SNP and SNPs with lower odds ratios, along with other risk factors, including age, gender, lifestyle habits, and environmental influences.

After we completed a new theory of discriminant analysis in 2015, we used Revised IP-Optimal Linear Discriminant Function (Revised IP-OLDF, RIP) to discriminate against the Alon microarray. The minimum number of misclassifications (Minimum NM, MNM) was zero, indicating that the data is linearly separable data (LSD). Two classes, noncancerous (“normal”) patients and cancer patients, are completely separable in the 2000 gene space. We found that LSD is a crucial signal for cancer gene diagnosis. We classify the linearly separable space and subspaces as Matryoshka. We found that LSD has two unique structures. First, LSD has a Matryoshka structure. LSD includes many smaller Matryoshkas in it up to the minimum Matryoshka (Basic Gene Set, BGS). The second structure is as follows: Program3 coded by LINGO can decompose LSD into the exclusive Small Matryoshkas (SMs) and other gene subspace (MNM > 0) by the Matryoshka Feature Selection Method (Method2). Program4 coded by LINGO can decompose LSD into the many BGSs and another gene set. The Program 4 algorithm follows the same procedure as the one used to find Yamanaka four genes from 24 genes. Just as iPS cells cannot generate when one of the four genes is deleted, deleting one gene from BGS does not make its MNM zero. Because the second structure releases us from the curse of high-dimensional data, we can analyze 64 SMs and 129 BGSs with JMP statistical software and propose a cancer gene diagnosis. Although we have successfully found all the signals, we need to rank the importance of all SMs and BGS for physicians to use for diagnosis. In this paper, we validate all 193 signals via the 100-fold cross-validation (Method1) and show the rank of all signals for the cancer diagnosis that is useful for medical research.

PubMed, the biomedical database of the National Institutes of Health, contains over 30 million abstracts. Comparisons between pairs of abstracts can reveal novel and highly valuable hypotheses, especially if the connected abstracts are in very different areas of biomedical knowledge. It is improbable that such hypotheses can be formulated exclusively by professional investigators, who are usually highly specialized. We developed an application that generates hypotheses by connecting pairs of facts: the first fact is contained in abstracts in the area of expertise or interest of an investigator, while the second fact can be anywhere in PubMed, inside or outside the investigator’s expertise. The application is based on a natural language processing machine learning model, provided by the Python package spaCy.

Next-generation sequencing provides a more efficient way to characterize the transcriptome; meanwhile, it introduces significant issues for the effective analysis of large-scale sequencing data. To counteract this growing problem, we developed a robust and modular pipeline for examining whole transcriptome sequencing data quickly. Utilizing a Docker for containerization and unanimous cross-platform support, we modularized the pipeline and created a practical, automated structure for pragmatic use. Furthermore, we have developed segregated implementations of the pipeline sections to allow end-users to craft custom pipelines easily. We applied our tool to various datasets and showed that our pipeline offered high-quality sequencing data analysis.

One way to improve classification performance and reliability is the combination of the decisions of multiple classifiers, which is usually known as late fusion. Late fusion has been applied in some biomedical applications, generally, using classic fusion methods, such as mean or majority voting. This work compares the performance of several state-of-the-art fusion methods on a novel biomedical application: automated stage classification of neuropsychological tests using electroencephalographic data. Those tests were made by epileptic patients to evaluate their memory and learning cognitive function with the following stages: stimulus display, retention interval, and subject response. The following late fusion methods were considered: Dempster-Shafer combination; alpha integration; copulas; independent component analysis mixture models; and behavior knowledge space. Late fusion was able to improve the performance of the single classifiers and the most stable results were achieved by alpha integration.

Computer-aided diagnosis of polyps has played an important role in advancing the screening capability of computed tomographic colonography (CTC). Texture features, including intensity, gradient, and curvature, are the essential components in differentiation of neoplastic and nonneoplastic polyps. Clinical study has shown that the malignancy rates of polyps increase in correlation to their size. In this chapter, we present a study to investigate the effect of separating polyps based on size on the performance of machine learning. First, the volume of interest of each polyp was extracted and further confirmed by Radiologists. All polyp masses in this study have a diameter size ranging from 6 to 30 mm. Then, we group polyps into three groups based on their sizes: 6–9 mm, 10–30 mm, and a combined group of 6–30 mm. The corresponding malignancy risks of polyps were also recorded. From each polyp volume, we extracted the traditional 14 Haralick texture features plus 16 additional features with a total 30 texture features. Those features were further grouped as descriptors for intensity, gradient, and curvature characteristics. Finally, we employed the Random Forest classifier to differentiate neoplastic and nonneoplastic polyps. The proposed texture feature analysis was studied using 228 polyp masses. We generated a measure of area under the curve (AUC) values of the receiver operating characteristics (ROC) curve. The performance was determined by assessing the sensitivity and specificity values. Experimental results demonstrated that gradient and curvature features were ideal for differentiation of the malignancy risk for medium-sized polyps, whereas intensity feature was better for smaller-sized polyps.

In this paper, a novel ECG classification algorithm based on long short-term memory (LSTM) and feature extraction was proposed to classify normal and abnormal ECG signals. We have taken RR interval, half width of QRS peak, as well as their distribution as the features of ECG signals. We trained the LSTM network with feature extraction to classify the imbalanced ECG signals. This research shows how to classify electrocardiogram (ECG) data from the PhysioNet using deep learning. The experimental results demonstrate that the proposed method achieves a good classification performance. The proposed method can be applied to assist cardiologists in more accurately and objectively diagnosing ECG signals.

From 1999 to 2004, six medical projects published papers and publicly uploaded microarrays on the Internet. We discriminated against the six data using Revised IP-OLDF (RIP) and found the six data’s minimum number of misclassifications (MNMs) are zero. This fact shows linearly separable classes in gene space and the six data are linearly separable data (LSD). We classify the linearly separable space and subspaces as Matryoshka. LSD is a crucial signal and has the following two vital structures: 1. LSD has the Matryoshka structure that includes smaller Matryoshkas (SM) up to the minimum dimensional SM (Basic Gene Set, BGS) in it. 2. LSD consists of many SMs or BGSs. This fact shows we are free from the curse of the high-dimensional microarray. Thus, we can analyze all SMs and BGSs and propose the cancer gene diagnosis. To confirm two truths, we discriminate 78 microarrays of 13 carcinomas collected from 2007 to 2017. Because we confirm that the six old and 78 new microarrays are LSD, we can open a new frontier for gene diagnosis, including microarrays and RNA-seq. Our theory is elementary for physicians. If physicians confirm whether their data are LSD, they can analyze all SMs and BGSs with statistical methods. We expect they use our methods, and they establish cancer diagnoses and better treat cancer patients.