Ali Akbari

Dr. Ali Akbari

Department of Biomedical Engineering, Texas A&M University, Galveston, TX 77843, USA

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
3 January 2020 in IEEE Transactions on Biomedical Engineering
Verified
Journal Article
Closed access
AAAli Akbari
Ali Akbari
RJRoozbeh Jafari
Roozbeh Jafari
Recognizing activities of daily living (ADL) provides vital contextual information that enhances the effectiveness of various mobile health and wellness applications. Development of wearable motion sensors along with machine learning algorithms offer a great opportunity for ADL recognition. However, the performance of the ADL recognition systems may significantly degrade when they are used by a new user due to inter-subject variability. This issue limits the usability of these systems. In this paper, we propose a deep learning assisted personalization framework for ADL recognition with the aim to maximize the personalization performance while minimizing solicitation of inputs or labels from the user to reduce user's burden. The proposed framework consists of unsupervised retraining of automatic feature extraction layers and supervised fine-tuning of classification layers through a novel active learning model based on a given model's uncertainty. We design a Bayesian deep convolutional neural network with stochastic latent variables that allows us to estimate both aleatoric (data-dependent) and epistemic (model-dependent) uncertainties in recognition task. In this study, for the first time, we show how distinguishing between the two aforementioned sources of uncertainty leads to more effective active learning. The experimental results show that our proposed method improves the accuracy of ADL recognition on a new user by 25% on average compared to the case of using a model for a new user with no personalization with an average final accuracy of 89.2%. Moreover, our method achieves higher personalization accuracy while significantly reducing user's burden in terms of soliciting inputs and labels compared to other methods.

27 July 2022 in Ethics and Information Technology
Verified
Journal Article
Closed access
PDF file
SASaleh Afroogh
Saleh Afroogh
AEAmir esmalian
Amir esmalian
AMAli Mostafavi
Ali Mostafavi
AAAli Akbari
Ali Akbari
KRKambiz Rasoulkhani
Kambiz Rasoulkhani
EHEhsan Hajiramezanali
Ehsan Hajiramezanali

11 January 2021 in ACM Transactions on Embedded Computing Systems
Verified
Journal Article
Open access
AAAli Akbari
Ali Akbari
JMJonathan Martinez
Jonathan Martinez
RJRoozbeh Jafari
Roozbeh Jafari
Annotating activities of daily living (ADL) is vital for developing machine learning models for activity recognition. In addition, it is critical for self-reporting purposes such as in assisted living where the users are asked to log their ADLs. However, data annotation becomes extremely challenging in real-world data collection scenarios, where the users have to provide annotations and labels on their own. Methods such as self-reports that rely on users’ memory and compliance are prone to human errors and become burdensome since they increase users’ cognitive load. In this article, we propose a light yet effective context-aware change point detection algorithm that is implemented and run on a smartwatch for facilitating data annotation for high-level ADLs. The proposed system detects the moments of transition from one to another activity and prompts the users to annotate their data. We leverage freely available Bluetooth low energy (BLE) information broadcasted by various devices to detect changes in environmental context. This contextual information is combined with a motion-based change point detection algorithm, which utilizes data from wearable motion sensors, to reduce the false positives and enhance the system's accuracy. Through real-world experiments, we show that the proposed system improves the quality and quantity of labels collected from users by reducing human errors while eliminating users’ cognitive load and facilitating the data annotation process.

16 August 2021 in IEEE Journal of Biomedical and Health Informatics
Verified
Journal Article
Closed access
AAAli Akbari
Ali Akbari
JMJonathan Martinez
Jonathan Martinez
RJRoozbeh Jafari
Roozbeh Jafari
Modality translation grants diagnostic value to wearable devices by translating signals collected from low-power sensors to their highly-interpretable counterparts that are more familiar to healthcare providers. For instance, bio-impedance (Bio-Z) is a conveniently collected modality for measuring physiological parameters but is not highly interpretable. Thus, translating it to a well-known modality such as electrocardiogram (ECG) improves the usability of Bio-Z in wearables. Deep learning solutions are well-suited for this task given complex relationships between modalities generated by distinct processes. However, current algorithms usually train a single model for all users that results in ignoring cross-user variations. Retraining for new users usually requires collecting abundant labeled data, which is challenging in healthcare applications. In this paper, we build a modality translation framework to translate Bio-Z to ECG by learning personalized user information without training several independent architectures. Furthermore, our framework is able to adapt to new users in testing using very few samples. We design a meta-learning framework that contains shared and user-specific parameters to account for user differences while learning from the similarity amongst user signals. In this model, a meta-learner approximated by a neural network learns how to learn user-specific parameters and can efficiently update them in testing. Our experiments show that the proposed model reduces the percentage root mean square difference (PRD) by 41% compared to training a single model for all users and by 36% compared to training independent models for each user. When adapting the model to new users, our model outperforms fine-tuning a pre-trained model through back-propagation by 40% using as few as two new samples in testing.

1 September 2018 in Medical Engineering & Physics
Verified
Journal Article
Closed access
Mohammad Iman Mokhlespour EsfahaniMohammad Iman Mokhlespour Esfahani
Mohammad Iman Mokhlespour Esfahani
AAAli Akbari
Ali Akbari
OZOmid Zobeiri
Omid Zobeiri
EREhsan Rashedi
Ehsan Rashedi
MPMohamad Parnianpour
Mohamad Parnianpour
The interest in wearable systems among the biomedical engineering and clinical community continues to escalate as technical refinements enhance their potential use for both indoor and outdoor applications. For example, an important wearable technology known as a microelectromechanical system (MEMS) is demonstrating promising applications in the area of biomedical engineering. Accordingly, this study was designed to investigate the Sharif-Human Movement Instrumentation System (SHARIF-HMIS), consisting of inertial measurement units (IMUs), stretchable clothing, and a data logger-all of which can be used outside the controlled environment of a laboratory, thus enhancing its overall utility. This system is lightweight, portable, able to be deliver data for almost 10 h, and features a new data-fusion algorithm using the Kalman filter with an adaptive approach. In specific terms, the data from the system's gyroscope, accelerometer, and magnetometer sensors can be combined to estimate total-body orientation; additionally, the noise level of these sensors can be changed to accommodate faster motions as well as magnetic disturbances. These variations can be incorporated within the extended Kalman filter by changing the parameters of the filter adaptively. In specific terms, the system's interface was developed to acquire data from eighteen IMUs located on the body to collect kinematic data associated with human motion. Meanwhile, a validation test involving one subject performing different shoulder motions was designed to compare data captured by SHARIF-HMIS and the VICON motion-capture system. This validation test demonstrated correlation values of >0.9. Results also confirmed that the output accuracy of the new system's sensor was <0.55, 1.5 and 3.5° for roll, pitch, and yaw directions, respectively. In summary, SHARIF-HMIS successfully collected kinematic data for specific human movements, which has promising implications for a range of sporting, biomedical, and healthcare-related applications.

10 September 2020 in IEEE Sensors Journal
Verified
Journal Article
Closed access
AAAli Akbari
Ali Akbari
RJRoozbeh Jafari
Roozbeh Jafari
Recognizing human daily activities with motion sensors data, specifically, modes of locomotion and transportation provides important contextual information that enhances the effectiveness of mobile applications. For instance, in assisted living or sports monitoring it is essential to log driving or running episodes. Previous studies in this field have utilized a fixed-size windowing technique for segmenting the sequential data of sensors. While segmenting signals into larger windows provides more information about the signal for classifiers, it increases misclassification rate when a transition occurs between the activities (i.e., multi-class windows). This will lead to inaccurate detection and logging of the activities of interest. To identify the exact time of transition from one to another activity as precisely as possible, this article proposes a fast and efficient hierarchical search algorithm that finds the exact sample at which transition occurs. This search algorithm can be applied to any activity recognition model with various lengths of segmentation window. To further improve the performance, we propose a new structure of 2D signal inputs to be used with 2D convolutional neural networks (CNN), which improves the ability of the CNN in capturing patterns underlying in inter-axes correlations. Experimental results show that the proposed transition detection method improves the F1-score by 28% compared to using fixed-size windowing approach for multi-class windows. In addition, the proposed method is 20 times faster than the naïve search.