1
|
Wang Q, Smythe D, Cao J, Hu Z, Proctor KJ, Owens AP, Zhao Y. Characterisation of Cognitive Load Using Machine Learning Classifiers of Electroencephalogram Data. Sensors (Basel) 2023; 23:8528. [PMID: 37896621 PMCID: PMC10611194 DOI: 10.3390/s23208528] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Received: 09/05/2023] [Revised: 10/09/2023] [Accepted: 10/16/2023] [Indexed: 10/29/2023]
Abstract
A high cognitive load can overload a person, potentially resulting in catastrophic accidents. It is therefore important to ensure the level of cognitive load associated with safety-critical tasks (such as driving a vehicle) remains manageable for drivers, enabling them to respond appropriately to changes in the driving environment. Although electroencephalography (EEG) has attracted significant interest in cognitive load research, few studies have used EEG to investigate cognitive load in the context of driving. This paper presents a feasibility study on the simulation of various levels of cognitive load through designing and implementing four driving tasks. We employ machine learning-based classification techniques using EEG recordings to differentiate driving conditions. An EEG dataset containing these four driving tasks from a group of 20 participants was collected to investigate whether EEG can be used as an indicator of changes in cognitive load. The collected dataset was used to train four Deep Neural Networks and four Support Vector Machine classification models. The results showed that the best model achieved a classification accuracy of 90.37%, utilising statistical features from multiple frequency bands in 24 EEG channels. Furthermore, the Gamma and Beta bands achieved higher classification accuracy than the Alpha and Theta bands during the analysis. The outcomes of this study have the potential to enhance the Human-Machine Interface of vehicles, contributing to improved safety.
Collapse
Affiliation(s)
- Qi Wang
- School of Aerospace, Transport and Manufacturing, Cranfield University, Cranfield MK43 0AL, UK; (Q.W.); (D.S.); (J.C.); (Z.H.)
| | - Daniel Smythe
- School of Aerospace, Transport and Manufacturing, Cranfield University, Cranfield MK43 0AL, UK; (Q.W.); (D.S.); (J.C.); (Z.H.)
| | - Jun Cao
- School of Aerospace, Transport and Manufacturing, Cranfield University, Cranfield MK43 0AL, UK; (Q.W.); (D.S.); (J.C.); (Z.H.)
| | - Zhilin Hu
- School of Aerospace, Transport and Manufacturing, Cranfield University, Cranfield MK43 0AL, UK; (Q.W.); (D.S.); (J.C.); (Z.H.)
| | - Karl J. Proctor
- Jaguar Land Rover Research, Coventry CV4 7AL, UK; (K.J.P.); (A.P.O.)
| | - Andrew P. Owens
- Jaguar Land Rover Research, Coventry CV4 7AL, UK; (K.J.P.); (A.P.O.)
| | - Yifan Zhao
- School of Aerospace, Transport and Manufacturing, Cranfield University, Cranfield MK43 0AL, UK; (Q.W.); (D.S.); (J.C.); (Z.H.)
| |
Collapse
|
2
|
Boyd A, Ye Z, Prabhu S, Tjong MC, Zha Y, Zapaishchykova A, Vajapeyam S, Hayat H, Chopra R, Liu KX, Nabavidazeh A, Resnick A, Mueller S, Haas-Kogan D, Aerts HJ, Poussaint T, Kann BH. Expert-level pediatric brain tumor segmentation in a limited data scenario with stepwise transfer learning. medRxiv 2023:2023.06.29.23292048. [PMID: 37425854 PMCID: PMC10327271 DOI: 10.1101/2023.06.29.23292048] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 07/11/2023]
Abstract
Purpose Artificial intelligence (AI)-automated tumor delineation for pediatric gliomas would enable real-time volumetric evaluation to support diagnosis, treatment response assessment, and clinical decision-making. Auto-segmentation algorithms for pediatric tumors are rare, due to limited data availability, and algorithms have yet to demonstrate clinical translation. Methods We leveraged two datasets from a national brain tumor consortium (n=184) and a pediatric cancer center (n=100) to develop, externally validate, and clinically benchmark deep learning neural networks for pediatric low-grade glioma (pLGG) segmentation using a novel in-domain, stepwise transfer learning approach. The best model [via Dice similarity coefficient (DSC)] was externally validated and subject to randomized, blinded evaluation by three expert clinicians wherein clinicians assessed clinical acceptability of expert- and AI-generated segmentations via 10-point Likert scales and Turing tests. Results The best AI model utilized in-domain, stepwise transfer learning (median DSC: 0.877 [IQR 0.715-0.914]) versus baseline model (median DSC 0.812 [IQR 0.559-0.888]; p<0.05). On external testing (n=60), the AI model yielded accuracy comparable to inter-expert agreement (median DSC: 0.834 [IQR 0.726-0.901] vs. 0.861 [IQR 0.795-0.905], p=0.13). On clinical benchmarking (n=100 scans, 300 segmentations from 3 experts), the experts rated the AI model higher on average compared to other experts (median Likert rating: 9 [IQR 7-9]) vs. 7 [IQR 7-9], p<0.05 for each). Additionally, the AI segmentations had significantly higher (p<0.05) overall acceptability compared to experts on average (80.2% vs. 65.4%). Experts correctly predicted the origins of AI segmentations in an average of 26.0% of cases. Conclusions Stepwise transfer learning enabled expert-level, automated pediatric brain tumor auto-segmentation and volumetric measurement with a high level of clinical acceptability. This approach may enable development and translation of AI imaging segmentation algorithms in limited data scenarios.
Collapse
Affiliation(s)
- Aidan Boyd
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Zezhong Ye
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Sanjay Prabhu
- Department of Radiology, Boston Children’s Hospital, Harvard Medical School, Boston, MA
| | - Michael C. Tjong
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Yining Zha
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Anna Zapaishchykova
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Sridhar Vajapeyam
- Department of Radiology, Boston Children’s Hospital, Harvard Medical School, Boston, MA
| | - Hasaan Hayat
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Rishi Chopra
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Kevin X. Liu
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Ali Nabavidazeh
- Center for Data-Driven Discovery in Biomedicine (D3b), Children’s Hospital of Philadelphia, Philadelphia, PA
- Department of Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA
| | - Adam Resnick
- Center for Data-Driven Discovery in Biomedicine (D3b), Children’s Hospital of Philadelphia, Philadelphia, PA
- Department of Neurosurgery, Children’s Hospital of Philadelphia, Philadelphia, PA
| | - Sabine Mueller
- Department of Neurology, University of California San Francisco, San Francisco, California
- Department of Pediatrics, University of California San Francisco, San Francisco, California
- Department of Neurological Surgery, University of California San Francisco, San Francisco, California
| | - Daphne Haas-Kogan
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| | - Hugo J.W.L. Aerts
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
- Department of Radiology, Brigham and Women’s Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, USA
- Radiology and Nuclear Medicine, CARIM & GROW, Maastricht University, Maastricht, the Netherlands
| | - Tina Poussaint
- Department of Radiology, Boston Children’s Hospital, Harvard Medical School, Boston, MA
| | - Benjamin H. Kann
- Artificial Intelligence in Medicine (AIM) Program, Mass General Brigham, Harvard Medical School, Boston, MA
- Department of Radiation Oncology, Dana-Farber Cancer Institute and Brigham and Women’s Hospital, Harvard Medical School, Boston, MA
| |
Collapse
|
3
|
Li X, Liu H, Chen X, Lyu Y, Liu Z. Inverse of initial stress in carbon fiber reinforced polymer laminates using lamb waves and deep neural network. Ultrasonics 2023; 132:107005. [PMID: 37043998 DOI: 10.1016/j.ultras.2023.107005] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 02/25/2022] [Revised: 03/31/2023] [Accepted: 04/02/2023] [Indexed: 05/29/2023]
Abstract
The prediction of the initial stress in composites is essential for the non-destructive testing (NDT) and structural health monitoring (SHM) of carbon fibre reinforced polymer (CFRP). This paper examines the potential of Lamb waves in the inverse of initial stress by calculating the influence of initial stress on the dispersion characteristics of Lamb waves propagating in multilayered CFRP laminates. By introducing the mechanics of incremental deformation into the linear three-dimensional elasticity theory, the Legendre orthogonal polynomial expansion (LOPE) method is used to mathematically model the Lamb wave propagating in multilayered CFRP laminates subjected to horizontal and vertical homogeneous initial stresses. Then, a three-hidden-layers Feed Forward Deep Neural Network (DNN) with Back Propagation (BP) algorithm is constructed to invert the magnitude and direction of the initial stresses. The input features are the phase velocities of fundamental Lamb wave A0 mode at five different frequencies. Both training and testing samples are obtained by LOPE forward calculation. An ablation experiment is presented to compare the two different activation functions. Finally, the accuracy of the inverse is verified by comparing with the available outcomes of LOPE forward calculation.
Collapse
Affiliation(s)
- Xuan Li
- School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Jungong Road 580, Shanghai 200093, PR China
| | - Hongye Liu
- School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Jungong Road 580, Shanghai 200093, PR China.
| | - Xin Chen
- Mechanical Engineering Division, Southwest Research Institute, 6220 Culebra Road, San Antonio, TX 78238, USA
| | - Yan Lyu
- Faculty of Materials and Manufacturing, Beijing University of Technology, Beijing 100124, PR China
| | - Zenghua Liu
- Faculty of Materials and Manufacturing, Beijing University of Technology, Beijing 100124, PR China
| |
Collapse
|
4
|
Wu X, Park S. A Prediction Model for Osteoporosis Risk Using a Machine-Learning Approach and Its Validation in a Large Cohort. J Korean Med Sci 2023; 38:e162. [PMID: 37270917 DOI: 10.3346/jkms.2023.38.e162] [Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 11/13/2022] [Accepted: 02/15/2023] [Indexed: 06/06/2023] Open
Abstract
BACKGROUND Osteoporosis develops in the elderly due to decreased bone mineral density (BMD), potentially increasing bone fracture risk. However, the BMD is not regularly measured in a clinical setting. This study aimed to develop a good prediction model for the osteoporosis risk using a machine learning (ML) approach in adults over 40 years in the Ansan/Anseong cohort and the association of predicted osteoporosis risk with a fracture in the Health Examinees (HEXA) cohort. METHODS The 109 demographic, anthropometric, biochemical, genetic, nutrient, and lifestyle variables of 8,842 participants were manually selected in an Ansan/Anseong cohort and included in the ML algorithm. The polygenic risk score (PRS) of osteoporosis was generated with a genome-wide association study and added for the genetic impact of osteoporosis. Osteoporosis was defined with < -2.5 T scores of the tibia or radius compared to people in their 20s-30s. They were divided randomly into the training (n = 7,074) and test (n = 1,768) sets-Pearson's correlation between the predicted osteoporosis risk and fracture in the HEXA cohort. RESULTS XGBoost, deep neural network, and random forest generated the prediction model with a high area under the curve (AUC, 0.86) of the receiver operating characteristic (ROC) with 10, 15, and 20 features; the prediction model by XGBoost had the highest AUC of ROC, high accuracy and k-fold values (> 0.85) in 15 features among seven ML approaches. The model included the genetic factor, genders, number of children and breastfed children, age, residence area, education, seasons to measure, height, smoking status, hormone replacement therapy, serum albumin, hip circumferences, vitamin B6 intake, and body weight. The prediction models for women alone were similar to those for both genders, with lower accuracy. When the prediction model was applied to the HEXA study, the correlation between the fracture incidence and predicted osteoporosis risk was significant but weak (r = 0.173, P < 0.001). CONCLUSION The prediction model for osteoporosis risk generated by XGBoost can be applied to estimate osteoporosis risk. The biomarkers can be considered for enhancing the prevention, detection, and early therapy of osteoporosis risk in Asians.
Collapse
Affiliation(s)
- Xuangao Wu
- Department of Bioconvergence, Hoseo University, Asan, Korea
| | - Sunmin Park
- Department of Bioconvergence, Hoseo University, Asan, Korea
- Department of Food and Nutrition, Obesity/Diabetes Research Center, Hoseo University, Asan, Korea.
| |
Collapse
|
5
|
Rajpal A, Sehra K, Bagri R, Sikka P. XAI-FR: Explainable AI-Based Face Recognition Using Deep Neural Networks. Wirel Pers Commun 2022; 129:663-680. [PMID: 36531522 PMCID: PMC9745692 DOI: 10.1007/s11277-022-10127-z] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Figures] [Subscribe] [Scholar Register] [Accepted: 11/30/2022] [Indexed: 06/17/2023]
Abstract
Face Recognition aims at identifying or confirming an individual's identity in a still image or video. Towards this end, machine learning and deep learning techniques have been successfully employed for face recognition. However, the response of the face recognition system often remains mysterious to the end-user. This paper aims to fill this gap by letting an end user know which features of the face has the model relied upon in recognizing a subject's face. In this context, we evaluate the interpretability of several face recognizers employing deep neural networks namely, LeNet-5, AlexNet, Inception-V3, and VGG16. For this purpose, a recently proposed explainable AI tool-Local Interpretable Model-Agnostic Explanations (LIME) is used. Benchmark datasets such as Yale, AT &T dataset, and Labeled Faces in the Wild (LFW) are utilized for this purpose. We are able to demonstrate that LIME indeed marks the features that are visually significant features for face recognition.
Collapse
Affiliation(s)
- Ankit Rajpal
- Department of Computer Science, University of Delhi, New Delhi, 110007 India
| | - Khushwant Sehra
- Department of Electronic Science, University of Delhi, South Campus, New Delhi, 110021 India
| | - Rashika Bagri
- Department of Computer Science, University of Delhi, New Delhi, 110007 India
| | - Pooja Sikka
- Department of Computer Science, University of Delhi, New Delhi, 110007 India
| |
Collapse
|
6
|
Wen S, Li X, Wang B, Tan J, Liu Y, Lv J, Tan Z, Yin L, Du Y. Deep Neural Network-Evaluated Thermal Conductivity for Two-Phase WC-M (M = Ag, Co) Cemented Carbides. Materials (Basel) 2022; 15:6269. [PMID: 36143580 PMCID: PMC9505479 DOI: 10.3390/ma15186269] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Figures] [Subscribe] [Scholar Register] [Received: 08/05/2022] [Revised: 08/27/2022] [Accepted: 09/07/2022] [Indexed: 06/16/2023]
Abstract
DNN (Deep Neural Network) is one kind of method for artificial intelligence, which has been applied in various fields including the exploration of material properties. In the present work, DNN, in combination with the 10-fold cross-validation, is applied to evaluate and predict the thermal conductivities for two-phase WC-M (M = Ag, Co) cemented carbides. Multi-layer DNNs were established by learning the measured thermal conductivities for the WC-Ag and WC-Co systems. It is observed that there are local-minimum regions for the loss functions during training and testing the DNNs, and the presently utilized Adam optimizer is valid for breaking the local-minimum regions. The good agreements between the DNN-evaluated thermal conductivities and the measured ones manifest that the DNNs were well trained and tested. Moreover, another 1000 input data points were randomly generated for the established DNNs to predict the thermal conductivities for WC-Ag and WC-Co systems, respectively. Compared with the thermal conductivities predicted by the previously developed physical model, the presently established DNNs show similarly robust predicting ability. Concerning the efficiency, it is demonstrated in the present work that machine learning is promising to explore the material properties, especially in the high-dimensional parameter space, more efficiently than previous models, and thus can considerably contribute to the corresponding material design with less time consumption and costs.
Collapse
Affiliation(s)
- Shiyi Wen
- School of Metallurgy and Environment, Central South University, Changsha 410083, China
| | - Xiaoguang Li
- Key Laboratory of Computing and Stochastic Mathematics (Ministry of Education), School of Mathematics and Statistics, Hunan Normal University, Changsha 410081, China
| | - Bo Wang
- Key Laboratory of Computing and Stochastic Mathematics (Ministry of Education), School of Mathematics and Statistics, Hunan Normal University, Changsha 410081, China
| | - Jing Tan
- State Key Laboratory of Powder Metallurgy, Central South University, Changsha 410083, China
| | - Yuling Liu
- State Key Laboratory of Powder Metallurgy, Central South University, Changsha 410083, China
| | - Jian Lv
- Institute of Engineering Research, Jiangxi University of Science and Technology, Ganzhou 341000, China
| | - Zhuopeng Tan
- Ganzhou Achteck Tool Technology Co., Ltd., Ganzhou 341000, China
| | - Lei Yin
- Ganzhou Achteck Tool Technology Co., Ltd., Ganzhou 341000, China
| | - Yong Du
- State Key Laboratory of Powder Metallurgy, Central South University, Changsha 410083, China
| |
Collapse
|
7
|
Haberbusch M, Bernardo LA, Galassi L, Oddo CM, Moscato F. Electrocardiogram Delineation Using Deep Neural Networks. Stud Health Technol Inform 2022; 293:117-118. [PMID: 35592969 DOI: 10.3233/shti220356] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 11/15/2022]
Abstract
BACKGROUND In recent years, there has been a rising interest in the application of deep neural networks (DNN) for the delineation of the electrocardiogram (ECG). OBJECTIVES A variety of DNN architectures has been investigated in a 5-fold cross-validation approach. RESULTS The best performing network achieved 100% sensitivity and >97% positive predictive value for all ECG waves. CONCLUSION Our DNN could achieve similar classification performance as other DNN approaches described in the literature at a reduced computational cost.
Collapse
Affiliation(s)
- Max Haberbusch
- Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria.,Ludwig Boltzmann Institute for Cardiovascular Engineering, Vienna, Austria
| | - Lisa A Bernardo
- Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria.,Scuola Superiore Sant'Anna, Pisa, Italy
| | - Laura Galassi
- Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria.,Scuola Superiore Sant'Anna, Pisa, Italy
| | - Calogero M Oddo
- Scuola Superiore Sant'Anna, Pisa, Italy.,The BioRobotics Institute, Scuola Superiore Sant'Anna, Pisa, Italy
| | - Francesco Moscato
- Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria.,Ludwig Boltzmann Institute for Cardiovascular Engineering, Vienna, Austria.,Austrian Cluster for Tissue Regeneration, Vienna, Austria
| |
Collapse
|
8
|
Liu D, Ding W, Dong ZS, Pedrycz W. Optimizing deep neural networks to predict the effect of social distancing on COVID-19 spread. Comput Ind Eng 2022; 166:107970. [PMID: 36568699 PMCID: PMC9757984 DOI: 10.1016/j.cie.2022.107970] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 06/17/2021] [Revised: 01/19/2022] [Accepted: 01/24/2022] [Indexed: 06/15/2023]
Abstract
Deep Neural Networks (DNN) form a powerful deep learning model that can process unprecedented volumes of data. The hyperparameters of DNN have a significant influence on its prediction performance. Evolutionary algorithms (EAs) form a heuristic-based approach that provides an opportunity to optimize deep learning models to obtain good performance. Therefore, we propose an evolutionary deep learning model called IPSO-DNN based on DNN for prediction and an improved Particle Swarm Optimization (IPSO) algorithm to optimize the kernel hyperparameters of DNN in a self-adaptive evolutionary way. In the IPSO algorithm, a micro population size setting is introduced to improve the search efficiency of the algorithm, and the generalized opposition-based learning strategy is used to guide the population evolution. In addition, the IPSO algorithm employs a self-adaptive update strategy to prevent premature convergence and then improves the exploitation and exploration parameter optimization performance of DNN. In this paper, we show that the IPSO algorithm provides an efficient approach for tuning the hyperparameters of DNN with saving valuable computational resources. We explore the proposed IPSO-DNN model to predict the effect of social distancing on the spread of COVID-19 based on the social distancing metrics. The preliminary experimental results reveal that the proposed IPSO-DNN model has the least computation cost and yields better prediction accuracy results when compared to the other models. The experiments of the IPSO-DNN model also illustrate that aggressive and extensive social distancing interventions are crucial to help flatten the COVID-19 epidemic curve in the United States.
Collapse
Affiliation(s)
- Dixizi Liu
- Department of Industrial Engineering, Clemson University, Clemson, SC 29634, United States
| | - Weiping Ding
- Centre for Artificial Intelligence, FEIT, University of Technology Sydney, Ultimo, NSW 2007, Austria
| | - Zhijie Sasha Dong
- Ingram School of Engineering, Texas State University, San Marcos, TX 78666, United States
| | - Witold Pedrycz
- Department of Electrical and Computer Engineering, University of Alberta, Edmonton, AB T6R 2V4, Canada
| |
Collapse
|
9
|
Nekooei A, Safari S. Compression of Deep Neural Networks based on quantized tensor decomposition to implement on reconfigurable hardware platforms. Neural Netw 2022; 150:350-363. [PMID: 35344706 DOI: 10.1016/j.neunet.2022.02.024] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/24/2021] [Revised: 11/28/2021] [Accepted: 02/24/2022] [Indexed: 11/24/2022]
Abstract
Deep Neural Networks (DNNs) have been vastly and successfully employed in various artificial intelligence and machine learning applications (e.g., image processing and natural language processing). As DNNs become deeper and enclose more filters per layer, they incur high computational costs and large memory consumption to preserve their large number of parameters. Moreover, present processing platforms (e.g., CPU, GPU, and FPGA) have not enough internal memory, and hence external memory storage is needed. Hence deploying DNNs on mobile applications is difficult, considering the limited storage space, computation power, energy supply, and real-time processing requirements. In this work, using a method based on tensor decomposition, network parameters were compressed, thereby reducing access to external memory. This compression method decomposes the network layers' weight tensor into a limited number of principal vectors such that (i) almost all the initial parameters can be retrieved, (ii) the network structure did not change, and (iii) the network quality after reproducing the parameters was almost similar to the original network in terms of detection accuracy. To optimize the realization of this method on FPGA, the tensor decomposition algorithm was modified while its convergence was not affected, and the reproduction of network parameters on FPGA was straightforward. The proposed algorithm reduced the parameters of ResNet50, VGG16, and VGG19 networks trained with Cifar10 and Cifar100 by almost 10 times.
Collapse
|
10
|
Shastri S, Kansal I, Kumar S, Singh K, Popli R, Mansotra V. CheXImageNet: a novel architecture for accurate classification of Covid-19 with chest x-ray digital images using deep convolutional neural networks. Health Technol (Berl) 2022; 12:193-204. [PMID: 35036283 PMCID: PMC8751458 DOI: 10.1007/s12553-021-00630-x] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/29/2021] [Accepted: 12/03/2021] [Indexed: 12/25/2022]
Abstract
Many countries around the world have been influenced by Covid-19 which is a serious virus as it gets transmitted by human communication. Although, its syndrome is quite similar to the ordinary flu. The critical step involved in Covid-19 is the initial screening or testing of the infected patients. As there are no special detection tools, the demand for such diagnostic tools has been increasing continuously. So, it is eminently admissible to find out positive cases of this disease at the earliest so that the spreading of this dangerous virus can be controlled. Although, some methods for the detection of Covid-19 patients are available, which are performed upon respiratory based samples and among them, a critical approach for treatment is radiologic imaging or X-ray imaging. The latest conclusions obtained from X-ray digital imaging based algorithms and techniques recommend that such type of digital images may consist of significant facts regarding the SARS-CoV-2 virus. The utilization of Deep Neural Networks based methodologies clubbed with digital radiological imaging has been proved useful for accurately identifying this disease. This could also be adjuvant in conquering the problem of dearth of competent physicians in far-flung areas. In this paper, a CheXImageNet model has been introduced for detecting Covid-19 disease by using digital images of Chest X-ray with the help of an openly accessible dataset. Experiments for both binary class and multi-class have been performed in this work for benchmarking the effectiveness of the proposed work. An accuracy of 100\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$\%$$\end{document}% is reported for both binary classification (having cases of Covid-19 and Normal X-Ray) and classification for three classes (including cases of Covid-19, Normal X-Ray and, cases of Pneumonia disease) respectively.
Collapse
Affiliation(s)
- Sourabh Shastri
- Department of Computer Science & IT, University of Jammu, 180006 Jammu & Kashmir, India
| | - Isha Kansal
- Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India
| | - Sachin Kumar
- Department of Computer Science & IT, University of Jammu, 180006 Jammu & Kashmir, India
| | - Kuljeet Singh
- Department of Computer Science & IT, University of Jammu, 180006 Jammu & Kashmir, India
| | - Renu Popli
- Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India
| | - Vibhakar Mansotra
- Department of Computer Science & IT, University of Jammu, 180006 Jammu & Kashmir, India
| |
Collapse
|
11
|
Naseem M, Arshad H, Hashmi SA, Irfan F, Ahmed FS. Predicting mortality in SARS-COV-2 (COVID-19) positive patients in the inpatient setting using a novel deep neural network. Int J Med Inform 2021; 154:104556. [PMID: 34455118 PMCID: PMC8378987 DOI: 10.1016/j.ijmedinf.2021.104556] [Citation(s) in RCA: 4] [Impact Index Per Article: 1.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/20/2021] [Revised: 08/03/2021] [Accepted: 08/15/2021] [Indexed: 02/07/2023]
Abstract
BACKGROUND The nextwave of COVID-19 pandemic is anticipated to be worse than the initial one and will strain the healthcare systems even more during the winter months. Our aim was to develop a novel machine learning-based model to predict mortality using the deep learning Neo-V framework. We hypothesized this novel machine learning approach could be applied to COVID-19 patients to predict mortality successfully with high accuracy. METHODS We collected clinical and laboratory data prospectively on all adult patients (≥18 years of age) that were admitted in the inpatient setting at Aga Khan University Hospital between February 2020 and September 2020 with a clinical diagnosis of COVID-19 infection. Only patients with a RT-PCR (reverse polymerase chain reaction) proven COVID-19 infection and complete medical records were included in this study. A Novel 3-phase machine learning framework was developed to predict mortality in the inpatients setting. Phase 1 included variable selection that was done using univariate and multivariate Cox-regression analysis; all variables that failed the regression analysis were excluded from the machine learning phase of the study. Phase 2 involved new-variables creation and selection. Phase 3 and final phase applied deep neural networks and other traditional machine learning models like Decision Tree Model, k-nearest neighbor models, etc. The accuracy of these models were evaluated using test-set accuracy, sensitivity, specificity, positive predictive values, negative predictive values and area under the receiver-operating curves. RESULTS After application of inclusion and exclusion criteria (n=)1214 patients were selected from a total of 1228 admitted patients. We observed that several clinical and laboratory-based variables were statistically significant for both univariate and multivariate analyses while others were not. With most significant being septic shock (hazard ratio [HR], 4.30; 95% confidence interval [CI], 2.91-6.37), supportive treatment (HR, 3.51; 95% CI, 2.01-6.14), abnormal international normalized ratio (INR) (HR, 3.24; 95% CI, 2.28-4.63), admission to the intensive care unit (ICU) (HR, 3.24; 95% CI, 2.22-4.74), treatment with invasive ventilation (HR, 3.21; 95% CI, 2.15-4.79) and laboratory lymphocytic derangement (HR, 2.79; 95% CI, 1.6-4.86). Machine learning results showed our deep neural network (DNN) (Neo-V) model outperformed all conventional machine learning models with test set accuracy of 99.53%, sensitivity of 89.87%, and specificity of 95.63%; positive predictive value, 50.00%; negative predictive value, 91.05%; and area under the receiver-operator curve of 88.5. CONCLUSION Our novel Deep-Neo-V model outperformed all other machine learning models. The model is easy to implement, user friendly and with high accuracy.
Collapse
Affiliation(s)
- Maleeha Naseem
- Department of Community Health Sciences, Aga Khan University, Karachi 74900, Pakistan
| | - Hajra Arshad
- Medical College, Aga Khan University, Karachi 74900, Pakistan
| | | | - Furqan Irfan
- College of Osteopathic Medicine, Institute of Global Health, Michigan State University, East Lansing, MI 48824, United States
| | - Fahad Shabbir Ahmed
- Clinicaro Machine Learning Group, New Haven, CT 06510, United States,Department of Pathology, Wayne State University, Detroit, MI 48201, United States,Corresponding author at: Department of Pathology, Wayne State University, Detroit, MI 48201, United States
| |
Collapse
|
12
|
Bender T, Seidler T, Bengel P, Sax U, Krefting D. Application of Pre-Trained Deep Learning Models for Clinical ECGs. Stud Health Technol Inform 2021; 283:39-45. [PMID: 34545818 DOI: 10.3233/SHTI210539] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register]
Abstract
Automatic electrocardiogram (ECG) analysis has been one of the very early use cases for computer assisted diagnosis (CAD). Most ECG devices provide some level of automatic ECG analysis. In the recent years, Deep Learning (DL) is increasingly used for this task, with the first models that claim to perform better than human physicians. In this manuscript, a pilot study is conducted to evaluate the added value of such a DL model to existing built-in analysis with respect to clinical relevance. 29 12-lead ECGs have been analyzed with a published DL model and results are compared to build-in analysis and clinical diagnosis. We could not reproduce the results of the test data exactly, presumably due to a different runtime environment. However, the errors were in the order of rounding errors and did not affect the final classification. The excellent performance in detection of left bundle branch block and atrial fibrillation that was reported in the publication could be reproduced. The DL method and the built-in method performed similarly good for the chosen cases regarding clinical relevance. While benefit of the DL method for research can be attested and usage in training can be envisioned, evaluation of added value in clinical practice would require a more comprehensive study with further and more complex cases.
Collapse
|
13
|
Zhao F, Wu Z, Wang L, Lin W, Xia S, Li G. A Deep Network for Joint Registration and Parcellation of Cortical Surfaces. Med Image Comput Comput Assist Interv 2021; 12904:171-181. [PMID: 35994035 PMCID: PMC9387764 DOI: 10.1007/978-3-030-87202-1_17] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/15/2023]
Abstract
Cortical surface registration and parcellation are two essential steps in neuroimaging analysis. Conventionally, they are performed independently as two tasks, ignoring the inherent connections of these two closely-related tasks. Essentially, both tasks rely on meaningful cortical feature representations, so they can be jointly optimized by learning shared useful cortical features. To this end, we propose a deep learning framework for joint cortical surface registration and parcellation. Specifically, our approach leverages the spherical topology of cortical surfaces and uses a spherical network as the shared encoder to first learn shared features for both tasks. Then we train two task-specific decoders for registration and parcellation, respectively. We further exploit the more explicit connection between them by incorporating the novel parcellation map similarity loss to enforce the boundary consistency of regions, thereby providing extra supervision for the registration task. Conversely, parcellation network training also benefits from the registration, which provides a large amount of augmented data by warping one surface with manual parcellation map to another surface, especially when only few manually-labeled surfaces are available. Experiments on a dataset with more than 600 cortical surfaces show that our approach achieves large improvements on both parcellation and registration accuracy (over separately trained networks) and enables training high-quality parcellation and registration models using much fewer labeled data.
Collapse
Affiliation(s)
- Fenqiang Zhao
- Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
| | - Zhengwang Wu
- Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
| | - Li Wang
- Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
| | - Weili Lin
- Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
| | - Shunren Xia
- Key Laboratory of Biomedical Engineering of Ministry of Education, Zhejiang University, Hangzhou, China
| | - Gang Li
- Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
| |
Collapse
|
14
|
Chen S, Wu S, Wang L. Hierarchical semantic interaction-based deep hashing network for cross-modal retrieval. PeerJ Comput Sci 2021; 7:e552. [PMID: 34141884 PMCID: PMC8176532 DOI: 10.7717/peerj-cs.552] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 03/12/2021] [Accepted: 04/28/2021] [Indexed: 06/12/2023]
Abstract
Due to the high efficiency of hashing technology and the high abstraction of deep networks, deep hashing has achieved appealing effectiveness and efficiency for large-scale cross-modal retrieval. However, how to efficiently measure the similarity of fine-grained multi-labels for multi-modal data and thoroughly explore the intermediate layers specific information of networks are still two challenges for high-performance cross-modal hashing retrieval. Thus, in this paper, we propose a novel Hierarchical Semantic Interaction-based Deep Hashing Network (HSIDHN) for large-scale cross-modal retrieval. In the proposed HSIDHN, the multi-scale and fusion operations are first applied to each layer of the network. A Bidirectional Bi-linear Interaction (BBI) policy is then designed to achieve the hierarchical semantic interaction among different layers, such that the capability of hash representations can be enhanced. Moreover, a dual-similarity measurement ("hard" similarity and "soft" similarity) is designed to calculate the semantic similarity of different modality data, aiming to better preserve the semantic correlation of multi-labels. Extensive experiment results on two large-scale public datasets have shown that the performance of our HSIDHN is competitive to state-of-the-art deep cross-modal hashing methods.
Collapse
Affiliation(s)
- Shubai Chen
- College of Computer and Information Science, Southwest University, Chongqing, People’s Republic of China
| | - Song Wu
- College of Computer and Information Science, Southwest University, Chongqing, People’s Republic of China
| | - Li Wang
- College of Electronic and Information Engineering, Southwest University, Chongqing, People’s Republic of China
| |
Collapse
|
15
|
Wang J, Zhang P, Li Y. Memory-Replay Knowledge Distillation. Sensors (Basel) 2021; 21:2792. [PMID: 33921068 DOI: 10.3390/s21082792] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 03/08/2021] [Revised: 04/06/2021] [Accepted: 04/09/2021] [Indexed: 11/24/2022]
Abstract
Knowledge Distillation (KD), which transfers the knowledge from a teacher to a student network by penalizing their Kullback–Leibler (KL) divergence, is a widely used tool for Deep Neural Network (DNN) compression in intelligent sensor systems. Traditional KD uses pre-trained teacher, while self-KD distills its own knowledge to achieve better performance. The role of the teacher in self-KD is usually played by multi-branch peers or the identical sample with different augmentation. However, the mentioned self-KD methods above have their limitation for widespread use. The former needs to redesign the DNN for different tasks, and the latter relies on the effectiveness of the augmentation method. To avoid the limitation above, we propose a new self-KD method, Memory-replay Knowledge Distillation (MrKD), that uses the historical models as teachers. Firstly, we propose a novel self-KD training method that penalizes the KD loss between the current model’s output distributions and its backup outputs on the training trajectory. This strategy can regularize the model with its historical output distribution space to stabilize the learning. Secondly, a simple Fully Connected Network (FCN) is applied to ensemble the historical teacher’s output for a better guidance. Finally, to ensure the teacher outputs offer the right class as ground truth, we correct the teacher logit output by the Knowledge Adjustment (KA) method. Experiments on the image (dataset CIFAR-100, CIFAR-10, and CINIC-10) and audio (dataset DCASE) classification tasks show that MrKD improves single model training and working efficiently across different datasets. In contrast to the existing fancy self-KD methods with various external knowledge, the effectiveness of MrKD sheds light on the usually abandoned historical models during the training trajectory.
Collapse
|
16
|
Kulkarni U, S M M, Gurlahosur SV, Bhogar G. Quantization Friendly MobileNet (QF-MobileNet) Architecture for Vision Based Applications on Embedded Platforms. Neural Netw 2021; 136:28-39. [PMID: 33429131 DOI: 10.1016/j.neunet.2020.12.022] [Citation(s) in RCA: 19] [Impact Index Per Article: 6.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 06/24/2020] [Revised: 11/10/2020] [Accepted: 12/23/2020] [Indexed: 11/17/2022]
Abstract
Deep Neural Networks (DNNs) have become popular for various applications in the domain of image and computer vision due to their well-established performance attributes. DNN algorithms involve powerful multilevel feature extractions resulting in an extensive range of parameters and memory footprints. However, memory bandwidth requirements, memory footprint and the associated power consumption of models are issues to be addressed to deploy DNN models on embedded platforms for real time vision-based applications. We present an optimized DNN model for memory and accuracy for vision-based applications on embedded platforms. In this paper we propose Quantization Friendly MobileNet (QF-MobileNet) architecture. The architecture is optimized for inference accuracy and reduced resource utilization. The optimization is obtained by addressing the redundancy and quantization loss of the existing baseline MobileNet architectures. We verify and validate the performance of the QF-MobileNet architecture for image classification task on the ImageNet dataset. The proposed model is tested for inference accuracy and resource utilization and compared to the baseline MobileNet architecture. The inference accuracy of the proposed QF-MobileNetV2 float model attained 73.36% and the quantized model has 69.51%. The MobileNetV3 float model attained an inference accuracy of 68.75% and the quantized model has 67.5% respectively. The proposed model saves 33% of time complexity for QF-MobileNetV2 and QF-MobileNetV3 models against the baseline models. The QF-MobileNet also showed optimized resource utilization with 32% fewer tunable parameters, 30% fewer MAC's operations per image and reduced inference quantization loss by approximately 5% compared to the baseline models. The model is ported onto the android application using TensorFlow API. The android application performs inference on the native devices viz. smartphones, tablets and handheld devices. Future work is focused on introducing channel-wise and layer-wise quantization schemes to the proposed model. We intend to explore quantization aware training of DNN algorithms to achieve optimized resource utilization and inference accuracy.
Collapse
Affiliation(s)
- Uday Kulkarni
- School of Computer Science & Engineering, KLE Technological University, Hubballi, India.
| | - Meena S M
- School of Computer Science & Engineering, KLE Technological University, Hubballi, India.
| | - Sunil V Gurlahosur
- School of Computer Science & Engineering, KLE Technological University, Hubballi, India.
| | - Gopal Bhogar
- School of Computer Science & Engineering, KLE Technological University, Hubballi, India.
| |
Collapse
|
17
|
Tong S, Liang X, Kumada T, Iwaki S. Putative ratios of facial attractiveness in a deep neural network. Vision Res 2020; 178:86-99. [PMID: 33186876 DOI: 10.1016/j.visres.2020.10.001] [Citation(s) in RCA: 6] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 11/12/2019] [Revised: 08/25/2020] [Accepted: 10/02/2020] [Indexed: 12/01/2022]
Abstract
Empirical evidence has shown that there is an ideal arrangement of facial features (ideal ratios) that can optimize the attractiveness of a person's face. These putative ratios define facial attractiveness in terms of spatial relations and provide important rules for measuring the attractiveness of a face. In this paper, we show that a deep neural network (DNN) model can learn putative ratios from face images based only on categorical annotation when no annotated facial features for attractiveness are explicitly given. To this end, we conducted three experiments. In Experiment 1, we trained a DNN model to recognize the attractiveness (female/male × high/low attractiveness) of face in the images using four category-specific neurons (CSNs). In Experiment 2, face-like images were generated by reversing the DNN model (e.g., deconvolution). These images depict the intuitive attributes encoded in CSNs of the four categories of facial attractiveness and reveal certain consistencies with reported evidence on the putative ratios. In Experiment 3, simulated psychophysical experiments on face images with varying putative ratios reveal changes in the activity of the CSNs that are remarkably similar to those of human judgements reported in a previous study. These results show that the trained DNN model can learn putative ratios as key features for the representation of facial attractiveness. This finding advances our understanding of facial attractiveness via DNN-based perspective approaches.
Collapse
Affiliation(s)
- Song Tong
- IST, Graduate School of Informatics, Kyoto University, Kyoto, Japan.
| | - Xuefeng Liang
- School of Artificial Intelligence, Xidian University, Xi'an, PR China.
| | - Takatsune Kumada
- IST, Graduate School of Informatics, Kyoto University, Kyoto, Japan.
| | - Sunao Iwaki
- Information Technology and Human Factors, AIST, Tsukuba, Japan.
| |
Collapse
|
18
|
Yoo H, Han S, Chung K. A Frequency Pattern Mining Model Based on Deep Neural Network for Real-Time Classification of Heart Conditions. Healthcare (Basel) 2020; 8:healthcare8030234. [PMID: 32722657 PMCID: PMC7551638 DOI: 10.3390/healthcare8030234] [Citation(s) in RCA: 13] [Impact Index Per Article: 3.3] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 05/30/2020] [Revised: 07/23/2020] [Accepted: 07/23/2020] [Indexed: 11/16/2022] Open
Abstract
Recently, a massive amount of big data of bioinformation is collected by sensor-based IoT devices. The collected data are also classified into different types of health big data in various techniques. A personalized analysis technique is a basis for judging the risk factors of personal cardiovascular disorders in real-time. The objective of this paper is to provide the model for the personalized heart condition classification in combination with the fast and effective preprocessing technique and deep neural network in order to process the real-time accumulated biosensor input data. The model can be useful to learn input data and develop an approximation function, and it can help users recognize risk situations. For the analysis of the pulse frequency, a fast Fourier transform is applied in preprocessing work. With the use of the frequency-by-frequency ratio data of the extracted power spectrum, data reduction is performed. To analyze the meanings of preprocessed data, a neural network algorithm is applied. In particular, a deep neural network is used to analyze and evaluate linear data. A deep neural network can make multiple layers and can establish an operation model of nodes with the use of gradient descent. The completed model was trained by classifying the ECG signals collected in advance into normal, control, and noise groups. Thereafter, the ECG signal input in real time through the trained deep neural network system was classified into normal, control, and noise. To evaluate the performance of the proposed model, this study utilized a ratio of data operation cost reduction and F-measure. As a result, with the use of fast Fourier transform and cumulative frequency percentage, the size of ECG reduced to 1:32. According to the analysis on the F-measure of the deep neural network, the model had 83.83% accuracy. Given the results, the modified deep neural network technique can reduce the size of big data in terms of computing work, and it is an effective system to reduce operation time.
Collapse
Affiliation(s)
- Hyun Yoo
- Department of Computer Engineering, Gachon University, Seongnam 13120, Korea;
| | - Soyoung Han
- Department of Nursing, Yonsei University Wonju College of Medicine, Wonju 26426, Korea;
| | - Kyungyong Chung
- Division of Computer Science and Engineering, Kyonggi University, Suwon 16227, Korea
- Correspondence:
| |
Collapse
|
19
|
Khan MN, Ahmed MM. Trajectory-level fog detection based on in-vehicle video camera with TensorFlow deep learning utilizing SHRP2 naturalistic driving data. Accid Anal Prev 2020; 142:105521. [PMID: 32408146 DOI: 10.1016/j.aap.2020.105521] [Citation(s) in RCA: 4] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 11/14/2019] [Revised: 03/07/2020] [Accepted: 03/22/2020] [Indexed: 06/11/2023]
Abstract
Providing drivers with real-time weather information and driving assistance during adverse weather, including fog, is crucial for safe driving. The primary focus of this study was to develop an affordable in-vehicle fog detection method, which will provide accurate trajectory-level weather information in real-time. The study used the SHRP2 Naturalistic Driving Study (NDS) video data and utilized several promising Deep Learning techniques, including Deep Neural Network (DNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN). Python programming on the TensorFlow Machine Learning library has been used for training the Deep Learning models. The analysis was done on a dataset consisted of three weather conditions, including clear, distant fog and near fog. During the training process, two optimizers, including Adam and Gradient Descent, have been used. While the overall prediction accuracy of the DNN, RNN, LSTM, and CNN using the Gradient Descent optimizer were found to be around 85 %, 77 %, 84 %, and 97 %, respectively; much improved overall prediction accuracy of 88 %, 91 %, 93 %, and 98 % for the DNN, RNN, LSTM, and CNN, respectively, were observed considering the Adam optimizer. The proposed fog detection method requires only a single video camera to detect weather conditions, and therefore, can be an inexpensive option to be fitted in maintenance vehicles to collect trajectory-level weather information in real-time for expanding as well as updating weather-based Variable Speed Limit (VSL) systems and Advanced Traveler Information Systems (ATIS).
Collapse
Affiliation(s)
- Md Nasim Khan
- University of Wyoming, Department of Civil & Architectural Engineering, 1000 E University Ave, Dept. 3295, Laramie, WY 82071, United States.
| | - Mohamed M Ahmed
- University of Wyoming, Department of Civil & Architectural Engineering, 1000 E University Ave, Dept. 3295, Laramie, WY 82071, United States.
| |
Collapse
|
20
|
Abstract
Children speech recognition is challenging mainly due to the inherent high variability in children's physical and articulatory characteristics and expressions. This variability manifests in both acoustic constructs and linguistic usage due to the rapidly changing developmental stage in children's life. Part of the challenge is due to the lack of large amounts of available children speech data for efficient modeling. This work attempts to address the key challenges using transfer learning from adult's models to children's models in a Deep Neural Network (DNN) framework for children's Automatic Speech Recognition (ASR) task evaluating on multiple children's speech corpora with a large vocabulary. The paper presents a systematic and an extensive analysis of the proposed transfer learning technique considering the key factors affecting children's speech recognition from prior literature. Evaluations are presented on (i) comparisons of earlier GMM-HMM and the newer DNN Models, (ii) effectiveness of standard adaptation techniques versus transfer learning, (iii) various adaptation configurations in tackling the variabilities present in children speech, in terms of (a) acoustic spectral variability, and (b) pronunciation variability and linguistic constraints. Our Analysis spans over (i) number of DNN model parameters (for adaptation), (ii) amount of adaptation data, (iii) ages of children, (iv) age dependent-independent adaptation. Finally, we provide Recommendations on (i) the favorable strategies over various aforementioned - analyzed parameters, and (ii) potential future research directions and relevant challenges/problems persisting in DNN based ASR for children's speech.
Collapse
Affiliation(s)
- Prashanth Gurunath Shivakumar
- Signal Processing for Communication Understanding & Behavior Analysis (SCUBA) Lab, University of Southern California, Los Angeles, California, USA
| | - Panayiotis Georgiou
- Signal Processing for Communication Understanding & Behavior Analysis (SCUBA) Lab, University of Southern California, Los Angeles, California, USA
| |
Collapse
|
21
|
Abstract
Since the QSAR/DNN model showed predominant predictive performance over other conventional methods in the Kaggle QSAR competition, many artificial neural network (ANN) methods have been applied to drug and material discovery. Appearance of artificial intelligence (AI), which is combined various general purpose ANN platforms with large-scale open access chemical databases, has attracting great interest and expectation in a wide range of molecular sciences. In this study, we investigate various DNN settings in order to reach a high-level of predictive performance comparable to the champion team of the competition, even with a general purpose ANN platform, and introduce the Meister setting for constructing a good QSAR/DNNs model. Here, we have used the most commonly available DNN model and constructed many QSAR/DNN models trained with various DNN settings by using the 15 datasets employed in the competition. As a result, it was confirmed that we can constructed the QSAR/DNN model that shows the same level of R2 performance as the champion team. The difference from the DNN setting recommended by the champion team was to reduce the mini-batch size. We have also explained that the R2 performance of each target depends on the molecular activity type, which is related to the complexity of biological mechanisms and chemical processes observed in molecular activity measurements.
Collapse
Affiliation(s)
- Yoshiki Kato
- Department of Computer Science and EngineeringToyohashi University of Technology1-1 Hibarigaoka, Tempaku choToyohashi, Aichi441-8580Japan
| | - Shinji Hamada
- Department of Computer Science and EngineeringToyohashi University of Technology1-1 Hibarigaoka, Tempaku choToyohashi, Aichi441-8580Japan
| | - Hitoshi Goto
- Department of Computer Science and EngineeringToyohashi University of Technology1-1 Hibarigaoka, Tempaku choToyohashi, Aichi441-8580Japan
| |
Collapse
|
22
|
Abdelhack M, Kamitani Y. Sharpening of Hierarchical Visual Feature Representations of Blurred Images. eNeuro 2018; 5:ENEURO. [PMID: 29756028 DOI: 10.1523/ENEURO.0443-17.2018] [Citation(s) in RCA: 9] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [What about the content of this article? (0)] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 12/19/2017] [Revised: 03/29/2018] [Accepted: 04/10/2018] [Indexed: 11/21/2022] Open
Abstract
The robustness of the visual system lies in its ability to perceive degraded images. This is achieved through interacting bottom-up, recurrent, and top-down pathways that process the visual input in concordance with stored prior information. The interaction mechanism by which they integrate visual input and prior information is still enigmatic. We present a new approach using deep neural network (DNN) representation to reveal the effects of such integration on degraded visual inputs. We transformed measured human brain activity resulting from viewing blurred images to the hierarchical representation space derived from a feedforward DNN. Transformed representations were found to veer toward the original nonblurred image and away from the blurred stimulus image. This indicated deblurring or sharpening in the neural representation, and possibly in our perception. We anticipate these results will help unravel the interplay mechanism between bottom-up, recurrent, and top-down pathways, leading to more comprehensive models of vision.
Collapse
|
23
|
Li X, Zhang Y, Li M, Marsic I, Yang J, Burd RS. Deep Neural Network for RFID-Based Activity Recognition. Proc Eighth Wirel Stud Stud Stud Workshop (2016) 2016; 2016:24-26. [PMID: 30506067 DOI: 10.1145/2987354.2987355] [Citation(s) in RCA: 3] [Impact Index Per Article: 0.4] [Reference Citation Analysis] [What about the content of this article? (0)] [Affiliation(s)] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 10/20/2022]
Abstract
We propose a Deep Neural Network (DNN) structure for RFID-based activity recognition. RFID data collected from several reader antennas with overlapping coverage have potential spatiotemporal relationships that can be used for object tracking. We augmented the standard fully-connected DNN structure with additional pooling layers to extract the most representative features. For model training and testing, we used RFID data from 12 tagged objects collected during 25 actual trauma resuscitations. Our results showed 76% recognition micro-accuracy for 7 resuscitation activities and 85% average micro-accuracy for 5 resuscitation phases, which is similar to existing system that, however, require the user to wear an RFID antenna.
Collapse
Affiliation(s)
- Xinyu Li
- Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ, USA
| | - Yanyi Zhang
- Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ, USA
| | - Mengzhu Li
- Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ, USA
| | - Ivan Marsic
- Department of Electrical and Computer Engineering, Rutgers University, Piscataway, NJ, USA
| | - JaeWon Yang
- Division of Trauma and Burn Surgery, Children's National Medical Center, Washington, D.C., USA
| | - Randall S Burd
- Division of Trauma and Burn Surgery, Children's National Medical Center, Washington, D.C., USA
| |
Collapse
|
24
|
Abstract
Alternative splicing significantly contributes to proteomic diversity and mis-regulation of splicing can cause diseases in human. Although both genomic and chromatin features have been shown to associate with splicing, the mechanisms by which various chromatin marks influence splicing is not clear for the most part. Moreover, it is not known whether the influence of specific genomic features on splicing is potentially modulated by the chromatin context. Here we report a deep neural network (DNN) model for predicting exon inclusion based on comprehensive genomic and chromatin features. Our analysis in three cell lines shows that, while both genomic and chromatin features can predict splicing to varying degrees, genomic features are the primary drivers of splicing, and the predictive power of chromatin features can largely be explained by their correlation with genomic features; chromatin features do not yield substantial independent contribution to splicing predictability. However, our model identified specific interactions between chromatin and genomic features suggesting that the effect of genomic elements may be modulated by chromatin context.
Collapse
Affiliation(s)
- Kun Wang
- Center for Bioinformatics and Computational Biology, University of Maryland
| | - Kan Cao
- Cell Biology Molecular Genetics, University of Maryland
| | | |
Collapse
|