This paper proposes a customized convolutional neural network for crack detection in concrete structures. The proposed method is compared to four existing deep learning methods based on training data size, data heterogeneity, network complexity, and the number of epochs. The performance of the proposed convolutional neural network (CNN) model is evaluated and compared to pretrained networks, i.e., the VGG-16, VGG-19, ResNet-50, and Inception V3 models, on eight datasets of different sizes, created from two public datasets. For each model, the evaluation considered computational time, crack localization results, and classification measures, e.g., accuracy, precision, recall, and F1-score. Experimental results demonstrated that training data size and heterogeneity among data samples significantly affect model performance. All models demonstrated promising performance on a limited number of diverse training data; however, increasing the training data size and reducing diversity reduced generalization performance, and led to overfitting. The proposed customized CNN and VGG-16 models outperformed the other methods in terms of classification, localization, and computational time on a small amount of data, and the results indicate that these two models demonstrate superior crack detection and localization for concrete structures.

Effective classroom instruction requires monitoring student participation and interaction during class, identifying cues to simulate their attention. The ability of teachers to analyze and evaluate students’ classroom behavior is becoming a crucial criterion for quality teaching. Artificial intelligence (AI)-based behavior recognition techniques can help evaluate students’ attention and engagement during classroom sessions. With rapid digitalization, the global education system is adapting and exploring emerging technological innovations, such as AI, the Internet of Things, and big data analytics, to improve education systems. In educational institutions, modern classroom systems are supplemented with the latest technologies to make them more interactive, student centered, and customized. However, it is difficult for instructors to assess students’ interest and attention levels even with these technologies. This study harnesses modern technology to introduce an intelligent real-time vision-based classroom to monitor students’ emotions, attendance, and attention levels even when they have face masks on. We used a machine learning approach to train students’ behavior recognition models, including identifying facial expressions, to identify students’ attention/non-attention in a classroom. The attention/no-attention dataset is collected based on nine categories. The dataset is given the YOLOv5 pre-trained weights for training. For validation, the performance of various versions of the YOLOv5 model (v5m, v5n, v5l, v5s, and v5x) are compared based on different evaluation measures (precision, recall, mAP, and F1 score). Our results show that all models show promising performance with 76% average accuracy. Applying the developed model can enable instructors to visualize students’ behavior and emotional states at different levels, allowing them to appropriately manage teaching sessions by considering student-centered learning scenarios. Overall, the proposed model will enhance instructors’ performance and students at an academic level.

This work introduces a predictive Length of Stay (LOS) framework for lung cancer patients using machine learning (ML) models. The framework proposed to deal with imbalanced datasets for classification-based approaches using electronic healthcare records (EHR). We have utilized supervised ML methods to predict lung cancer inpatients LOS during ICU hospitalization using the MIMIC-III dataset. Random Forest (RF) Model outperformed other models and achieved predicted results during the three framework phases. With clinical significance features selection, over-sampling methods (SMOTE and ADASYN) achieved the highest AUC results (98% with CI 95%: 95.3–100%, and 100% respectively). The combination of Over-sampling and under-sampling achieved the second-highest AUC results (98%, with CI 95%: 95.3–100%, and 97%, CI 95%: 93.7–100% SMOTE-Tomek, and SMOTE-ENN respectively). Under-sampling methods reported the least important AUC results (50%, with CI 95%: 40.2–59.8%) for both (ENN and Tomek- Links). Using ML explainable technique called SHAP, we explained the outcome of the predictive model (RF) with SMOTE class balancing technique to understand the most significant clinical features that contributed to predicting lung cancer LOS with the RF model. Our promising framework allows us to employ ML techniques in-hospital clinical information systems to predict lung cancer admissions into ICU.

Background: Robot-assisted therapy has become a promising technology in the field of rehabilitation for poststroke patients with motor disorders. Motivation during the rehabilitation process is a top priority for most stroke survivors. With current advancements in technology there has been the introduction of virtual reality (VR), augmented reality (AR), customizable games, or a combination thereof, that aid robotic therapy in retaining, or increasing the interests of, patients so they keep performing their exercises. However, there are gaps in the evidence regarding the transition from clinical rehabilitation to home-based therapy which calls for an updated synthesis of the literature that showcases this trend. The present review proposes a categorization of these studies according to technologies used, and details research in both upper limb and lower limb applications. Objective: The goal of this work was to review the practices and technologies implemented in the rehabilitation of poststroke patients. It aims to assess the effectiveness of exoskeleton robotics in conjunction with any of the three technologies (VR, AR, or gamification) in improving activity and participation in poststroke survivors. Methods: A systematic search of the literature on exoskeleton robotics applied with any of the three technologies of interest (VR, AR, or gamification) was performed in the following databases: MEDLINE, EMBASE, Science Direct & The Cochrane Library. Exoskeleton-based studies that did not include any VR, AR or gamification elements were excluded, but publications from the years 2010 to 2017 were included. Results in the form of improvements in the patients’ condition were also recorded and taken into consideration in determining the effectiveness of any of the therapies on the patients. Results: Thirty studies were identified based on the inclusion criteria, and this included randomized controlled trials as well as exploratory research pieces. There were a total of about 385 participants across the various studies. The use of technologies such as VR-, AR-, or gamification-based exoskeletons could fill the transition from the clinic to a home-based setting. Our analysis showed that there were general improvements in the motor function of patients using the novel interfacing techniques with exoskeletons. This categorization of studies helps with understanding the scope of rehabilitation therapies that can be successfully arranged for home-based rehabilitation. Conclusions: Future studies are necessary to explore various types of customizable games required to retain or increase the motivation of patients going through the individual therapies.

Elderly population (over the age of 60) is predicted to be 1.2 billion by 2025. Most of the elderly people would like to stay alone in their own house due to the high eldercare cost and privacy invasion. Unobtrusive activity recognition is the most preferred solution for monitoring daily activities of the elderly people living alone rather than the camera and wearable devices based systems. Thus, we propose an unobtrusive activity recognition classifier using Deep Convolutional Neural Network (DCNN) and anonymous binary sensors (PIR sensors and door sensors). The real world long-term fully annotated Aruba open dataset collected by binary sensors has been employed for the training and evaluation of the classifier. First, ten basic daily activities, namely: Bed_to_Toilet, Eating, Meal_Preparation, Relax, Sleeping, Work, Housekeeping, Wash_Dishes, Enter_Home, and Leave_Home are segmented with different sliding window sizes, and then converted into binary activity images. Next, the activity images are employed for training the DCNN classifiers with different parameters. Finally, the trained classifiers are evaluated with the 10-fold cross validation method, and the results show that the best DCNN classifier gives 0.79 and 0.951 of F1-score for all 10 activities and eight activities (excluding Leave_Home and Wash_Dishes), respectively.