Ali Khadem
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ali Khadem has a total of 66 co-authors. Here are the top co-authors by number of shared publications:
- Afshin Shoeibi
Data Science and Computational Intelligence Institute, University of Granada, Spain
3 shared publications Abbas Khosravi
Institute for Intelligent Systems Research and Innovation (IISRI), Deakin University, Victoria, 3217, Australia
2 shared publicationsAssef Zare
Faculty of Electrical Engineering, Gonabad Branch, Islamic Azad University, Gonabad, Iran
2 shared publications- Erfan Ebrahim Esfahani
Department of Biomedical Engineering, Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran
2 shared publications - Hamid Abrishami Moghaddam
Department of Biomedical Engineering, Faculty of Electrical Engineering, K. N. Toosi University of Technology, Tehran, Iran
2 shared publications Mani Garousi
Department of Biomedical Engineering, Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran
2 shared publications
Latest Publications
- Open Access
- 23 August 2026
- Published by Springer Nature in Scientific Reports
A hybrid 3D-CNN and CGAN approach for bipolar disorder diagnosis using VBM and SBC features from multimodal MRI
- 23 August 2026
- Published by Springer Nature in Scientific Reports
Open AccessA hybrid 3D-CNN and CGAN approach for bipolar disorder diagnosis using VBM and SBC features from multimodal MRI
- 3 February 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 118
Deep learning for diagnosing autism spectrum disorder using EEG with upright and inverted face processing tasks
- 3 February 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 118
Deep learning for diagnosing autism spectrum disorder using EEG with upright and inverted face processing tasks
- Open Access
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 206721-206733
Action Potential Restoration by Deep Learning on Sparsifying Transforms for Compressive Neural Recording in Brain Implants
Implantable brain chips record neural action potentials from hundreds of channels. Acquisition and wireless transmission of this data usually mandates high silicon area and power consumption, both potentially compromising practicality and patient safety in chronic use. To mitigate such complexities, on-chip compression of recorded action potentials based on sparsifying transforms has long been suggested as a possible solution. Nevertheless, due to the chip limitations pointed out, only a handful of largest-magnitude transform coefficients are allowed to be transmitted, inevitably degrading signal reconstruction on the external host. In order to improve restoration quality, we propose to complement the reconstruction process on the external side following the inverse transform with a deep lightweight convolutional neural network (CNN) trained to map the partial reconstruction to a more faithful one. This way, we can obtain significantly better results on the external side without ever increasing the number of transmitted coefficients, computational load or power and area take-up of the on-chip transform. colorblueSplitting the benchmark hc-1 data by 10 training, 10 test and 4 validation rats, experiments with the proposed method based on well-established on-chip transforms such as symmlet-4 wavelet (Sym4) and Walsh-Hadamard transform (WHT) indicate notable improvements in mean signal-to-noise-and-distortion ratios (SNDR) at minimal hardware costs. For instance, at 8-fold compression (N/M = 64/8), the proposed method based on WHT (i.e. WHT-CNN) improves mean SNDR from 7.81 dB (PCA-based method from 2025) to 9.07 dB, see Table 4. The normalized hardware cost for WHT-CNN is 0.28 ÎĽW power consumption and 0.0067mm2 area take-up per channel, while for PCA, the figures are 4.39 ÎĽW and 0.028mm2, respectively, where normalization means projecting the hardware to 180 nm CMOS technology and 1.2 V supply voltage. Moreover, the lightweight nature of our approach allows for it to deployed on low-resource devices such as smartphones. Our code and data are available at https://github.com/ErfanEbEsf/NSP3.
- 3 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 206721-206733
Open AccessAction Potential Restoration by Deep Learning on Sparsifying Transforms for Compressive Neural Recording in Brain Implants
Implantable brain chips record neural action potentials from hundreds of channels. Acquisition and wireless transmission of this data usually mandates high silicon area and power consumption, both potentially compromising practicality and patient safety in chronic use. To mitigate such complexities, on-chip compression of recorded action potentials based on sparsifying transforms has long been suggested as a possible solution. Nevertheless, due to the chip limitations pointed out, only a handful of largest-magnitude transform coefficients are allowed to be transmitted, inevitably degrading signal reconstruction on the external host. In order to improve restoration quality, we propose to complement the reconstruction process on the external side following the inverse transform with a deep lightweight convolutional neural network (CNN) trained to map the partial reconstruction to a more faithful one. This way, we can obtain significantly better results on the external side without ever increasing the number of transmitted coefficients, computational load or power and area take-up of the on-chip transform. colorblueSplitting the benchmark hc-1 data by 10 training, 10 test and 4 validation rats, experiments with the proposed method based on well-established on-chip transforms such as symmlet-4 wavelet (Sym4) and Walsh-Hadamard transform (WHT) indicate notable improvements in mean signal-to-noise-and-distortion ratios (SNDR) at minimal hardware costs. For instance, at 8-fold compression (N/M = 64/8), the proposed method based on WHT (i.e. WHT-CNN) improves mean SNDR from 7.81 dB (PCA-based method from 2025) to 9.07 dB, see Table 4. The normalized hardware cost for WHT-CNN is 0.28 ÎĽW power consumption and 0.0067mm2 area take-up per channel, while for PCA, the figures are 4.39 ÎĽW and 0.028mm2, respectively, where normalization means projecting the hardware to 180 nm CMOS technology and 1.2 V supply voltage. Moreover, the lightweight nature of our approach allows for it to deployed on low-resource devices such as smartphones. Our code and data are available at https://github.com/ErfanEbEsf/NSP3.
- 27 October 2025
- Published by Springer Nature in Brain Structure and Function
- Vol. 230  (8) , 1-14
Assessment of resting state structural-functional relationships in perisylvian region during the early weeks after birth
- 27 October 2025
- Published by Springer Nature in Brain Structure and Function
- Vol. 230  (8) , 1-14
Assessment of resting state structural-functional relationships in perisylvian region during the early weeks after birth
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. 55  (10) , 7228-7237
Arithmetic-Free Personalized Compressed Sensing Based On Deep Neural Networks for Wireless Transmission From Brain–Computer Interfaces
State-of-the-art brain–computer interfaces can carry out neural recording from hundreds of channels with high resolution. Such massive data makes it easy to study the brain better than ever before, but on the flip side, it leads to increased chip size, power consumption, heat dissipation and risk for patient safety. As such, compression of the data prior to transmission from the implant could be key to improving reliability and usability of such microsystems. In recent years, starting from sparsifying transforms all the way to compressive autoencoders (AEs), this compression has been offered by substantial arithmetic on the implant side, which in turn incurs its own inevitable costs. In this work, we analyze spike waveforms to prioritize subintervals by their importance. Thereupon, we design a temporal undersampling pattern matching the importance of each subinterval for compressive sensing of spikes. Following such sensing, we reconstruct spikes using a deep neural network (DNN) trained to capture spike representation from the undersampled measurements, with possible adaptation to individual subjects. This approach offers what we believe is the first spike compression-reconstruction framework that imposes no arithmetic on the compressing side, yet on the restoration side, performs at least on par with most on-chip arithmetic-heavy techniques. For instance, given a spike length of N=64 at eightfold compression, the famed symmlet-4 method yields a mean signal-to-noise-and-distortion ratio (SNDR) of 7.14 dB at a total compression arithmetic cost of 16N sums and products per spike, while for the proposed method, the figure is 8.38 dB at 0 sums and products.
- 22 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Systems, Man, and Cybernetics: Systems
- Vol. 55  (10) , 7228-7237
Arithmetic-Free Personalized Compressed Sensing Based On Deep Neural Networks for Wireless Transmission From Brain–Computer Interfaces
State-of-the-art brain–computer interfaces can carry out neural recording from hundreds of channels with high resolution. Such massive data makes it easy to study the brain better than ever before, but on the flip side, it leads to increased chip size, power consumption, heat dissipation and risk for patient safety. As such, compression of the data prior to transmission from the implant could be key to improving reliability and usability of such microsystems. In recent years, starting from sparsifying transforms all the way to compressive autoencoders (AEs), this compression has been offered by substantial arithmetic on the implant side, which in turn incurs its own inevitable costs. In this work, we analyze spike waveforms to prioritize subintervals by their importance. Thereupon, we design a temporal undersampling pattern matching the importance of each subinterval for compressive sensing of spikes. Following such sensing, we reconstruct spikes using a deep neural network (DNN) trained to capture spike representation from the undersampled measurements, with possible adaptation to individual subjects. This approach offers what we believe is the first spike compression-reconstruction framework that imposes no arithmetic on the compressing side, yet on the restoration side, performs at least on par with most on-chip arithmetic-heavy techniques. For instance, given a spike length of N=64 at eightfold compression, the famed symmlet-4 method yields a mean signal-to-noise-and-distortion ratio (SNDR) of 7.14 dB at a total compression arithmetic cost of 16N sums and products per spike, while for the proposed method, the figure is 8.38 dB at 0 sums and products.
- Open Access
- 1 January 2025
- Published by Salvia Medical Sciences Ltd in Journal of Biomedical Physics and Engineering
- Vol. online  (3) , 231-244
Diagnosing Epilepsy using Entropy Measures and Embedding Parameters of EEG Signals
Background: Manual analysis of electroencephalography (EEG) for epilepsy diagnosis can be subjective and time-consuming, leading to potential errors. An automatic classification system with high detection accuracy is essential for improving diagnostic efficiency and reliability.Objective: This study aimed to evaluate a comprehensive set of entropy measures, along with embedding parameters, to identify the most effective single measure for epilepsy diagnosis.Material and Methods: This analytical study used EEG data from the University of Bonn, including healthy controls (HCs) with open eyes and epileptic seizure patients, each with 100 single-channel segments. Discrete wavelet transform was applied, extracting ten entropy measures and two embedding parameters. Statistical tests evaluated feature significance, and a linear discriminant analysis (LDA) classifier was used for classification. Robustness was assessed by introducing Gaussian noise at varying signal-to-noise ratios (SNRs) and analyzing classification performance.Results: Our findings indicated that embedding parameters, permutation entropy, fuzzy entropy, sample entropy, norm entropy, sure entropy, log entropy, and threshold entropy significantly differentiated epileptic patients from HCs. Among these, sample entropy, norm entropy, sure entropy, log entropy, threshold entropy, and embedding delay achieved classification accuracies between 97% and 100% using LDA classifier. Furthermore, even with substantial Gaussian noise, the classifier maintained an accuracy above 84%, demonstrating the robustness of these features in noisy conditions. Conclusion: This study demonstrated that embedding-based and entropy-based features can serve as effective individual measures for discriminating epileptic EEG signals from HCs. These findings underscore the potential of these measures in automated epilepsy diagnosis systems, resulting in a robust and reliable tool for clinical applications.
- 1 January 2025
- Published by Salvia Medical Sciences Ltd in Journal of Biomedical Physics and Engineering
- Vol. online  (3) , 231-244
Open AccessDiagnosing Epilepsy using Entropy Measures and Embedding Parameters of EEG Signals
Background: Manual analysis of electroencephalography (EEG) for epilepsy diagnosis can be subjective and time-consuming, leading to potential errors. An automatic classification system with high detection accuracy is essential for improving diagnostic efficiency and reliability.Objective: This study aimed to evaluate a comprehensive set of entropy measures, along with embedding parameters, to identify the most effective single measure for epilepsy diagnosis.Material and Methods: This analytical study used EEG data from the University of Bonn, including healthy controls (HCs) with open eyes and epileptic seizure patients, each with 100 single-channel segments. Discrete wavelet transform was applied, extracting ten entropy measures and two embedding parameters. Statistical tests evaluated feature significance, and a linear discriminant analysis (LDA) classifier was used for classification. Robustness was assessed by introducing Gaussian noise at varying signal-to-noise ratios (SNRs) and analyzing classification performance.Results: Our findings indicated that embedding parameters, permutation entropy, fuzzy entropy, sample entropy, norm entropy, sure entropy, log entropy, and threshold entropy significantly differentiated epileptic patients from HCs. Among these, sample entropy, norm entropy, sure entropy, log entropy, threshold entropy, and embedding delay achieved classification accuracies between 97% and 100% using LDA classifier. Furthermore, even with substantial Gaussian noise, the classifier maintained an accuracy above 84%, demonstrating the robustness of these features in noisy conditions. Conclusion: This study demonstrated that embedding-based and entropy-based features can serve as effective individual measures for discriminating epileptic EEG signals from HCs. These findings underscore the potential of these measures in automated epilepsy diagnosis systems, resulting in a robust and reliable tool for clinical applications.
- 3 December 2024
- Published by SAGE Publications in Journal of Alzheimer’s Disease
- Vol. 103  (2) , 452-464
Classifying cognitive impairment based on FDG-PET and combined T1-MRI and rs-fMRI: An ADNI study
Background: Mild cognitive impairment (MCI) refers to a memory impairment among non-demented adults. It is a condition that increases the risk of dementia, notably due to Alzheimer's disease (AD). MCI is heterogeneous and there is a need for novel diagnostic approaches. Fluorodeoxyglucose positron emission tomography (FDG-PET) imaging provides robust AD biomarker characteristics, while anatomical and functional magnetic resonance imaging (MRI) offer complementary information. Objective: Classify MCI and cognitively normal (CN) adults using FDG-PET images; predict individuals with MCI that convert to AD dementia; determine if MRI can achieve comparable performance to FDG-PET classification. Methods: Four ADNI cohorts were created. Cohort 1: 805 participants (MCI n = 455; CN n = 350) that underwent FDG-PET. FDG-PET images were inputs to a one-channel 3-dimensional (3D) DenseNet deep learning model. Cohort 2: 348 participants (MCI n = 174; CN n = 174) with MRI and functional MRI. Cohort 3: overlapping cases from cohorts 1 and 2 (MCI n = 70; CN n = 70). Cohort 4: 336 participants (MCI-converters n = 168; MCI-stable n = 168) with FDG-PET from cohort 1. The one/two-channel models’ inputs were T1-weighted MRI and/or amplitude of low-frequency fluctuations images, with classification metrics evaluated through 10-fold cross-validation. Results: The FDG-PET model achieved 88.02%±3.82 accuracy for MCI versus CN classification, with 88.70%±4.70 sensitivity and 87.14%±5.03 specificity. Neither MRI model outperformed the FDG-PET model, as the highest MRI-based accuracy was 76.86%±1.95. The FDG-PET model achieved 63.23%±4.68 accuracy in classifying MCI-converters versus MCI-stable. Conclusions: FDG-PET images produced the highest accuracy in classifying MCI versus CN. While MRI-based approaches were inferior to FDG-PET, multi-contrast MRI still offers value for neurodegeneration classification.
- 3 December 2024
- Published by SAGE Publications in Journal of Alzheimer’s Disease
- Vol. 103  (2) , 452-464
Classifying cognitive impairment based on FDG-PET and combined T1-MRI and rs-fMRI: An ADNI study
Background: Mild cognitive impairment (MCI) refers to a memory impairment among non-demented adults. It is a condition that increases the risk of dementia, notably due to Alzheimer's disease (AD). MCI is heterogeneous and there is a need for novel diagnostic approaches. Fluorodeoxyglucose positron emission tomography (FDG-PET) imaging provides robust AD biomarker characteristics, while anatomical and functional magnetic resonance imaging (MRI) offer complementary information. Objective: Classify MCI and cognitively normal (CN) adults using FDG-PET images; predict individuals with MCI that convert to AD dementia; determine if MRI can achieve comparable performance to FDG-PET classification. Methods: Four ADNI cohorts were created. Cohort 1: 805 participants (MCI n = 455; CN n = 350) that underwent FDG-PET. FDG-PET images were inputs to a one-channel 3-dimensional (3D) DenseNet deep learning model. Cohort 2: 348 participants (MCI n = 174; CN n = 174) with MRI and functional MRI. Cohort 3: overlapping cases from cohorts 1 and 2 (MCI n = 70; CN n = 70). Cohort 4: 336 participants (MCI-converters n = 168; MCI-stable n = 168) with FDG-PET from cohort 1. The one/two-channel models’ inputs were T1-weighted MRI and/or amplitude of low-frequency fluctuations images, with classification metrics evaluated through 10-fold cross-validation. Results: The FDG-PET model achieved 88.02%±3.82 accuracy for MCI versus CN classification, with 88.70%±4.70 sensitivity and 87.14%±5.03 specificity. Neither MRI model outperformed the FDG-PET model, as the highest MRI-based accuracy was 76.86%±1.95. The FDG-PET model achieved 63.23%±4.68 accuracy in classifying MCI-converters versus MCI-stable. Conclusions: FDG-PET images produced the highest accuracy in classifying MCI versus CN. While MRI-based approaches were inferior to FDG-PET, multi-contrast MRI still offers value for neurodegeneration classification.
- Open Access
- 29 August 2024
- Published by Springer Nature in Scientific Reports
- Vol. 14  (1) , 1-10
Twinned neuroimaging analysis contributes to improving the classification of young people with autism spectrum disorder
Autism spectrum disorder (ASD) is diagnosed using comprehensive behavioral information. Neuroimaging offers additional information but lacks clinical utility for diagnosis. This study investigates whether multi-forms of magnetic resonance imaging (MRI) contrast can be used individually and in combination to produce a categorical classification of young individuals with ASD. MRI data were accessed from the Autism Brain Imaging Data Exchange (ABIDE). Young participants (ages 2–30) were selected, and two group cohorts consisted of 702 participants: 351 ASD and 351 controls. Image-based classification was performed using one-channel and two-channel inputs to 3D-DenseNet deep learning networks. The models were trained and tested using tenfold cross-validation. Two-channel models were twinned with combinations of structural MRI (sMRI) maps and amplitude of low-frequency fluctuations (ALFF) or fractional ALFF (fALFF) maps from resting-state functional MRI (rs-fMRI). All models produced classification accuracy that exceeded 65.1%. The two-channel ALFF-sMRI model achieved the highest mean accuracy of 76.9% ± 2.34. The one-channel ALFF-based model alone had mean accuracy of 72% ± 3.1. This study leveraged the ABIDE dataset to produce ASD classification results that are comparable and/or exceed literature values. The deep learning approach was conducive to diverse neuroimaging inputs. Findings reveal that the ALFF-sMRI two-channel model outperformed all others.
- 29 August 2024
- Published by Springer Nature in Scientific Reports
- Vol. 14  (1) , 1-10
Open AccessTwinned neuroimaging analysis contributes to improving the classification of young people with autism spectrum disorder
Autism spectrum disorder (ASD) is diagnosed using comprehensive behavioral information. Neuroimaging offers additional information but lacks clinical utility for diagnosis. This study investigates whether multi-forms of magnetic resonance imaging (MRI) contrast can be used individually and in combination to produce a categorical classification of young individuals with ASD. MRI data were accessed from the Autism Brain Imaging Data Exchange (ABIDE). Young participants (ages 2–30) were selected, and two group cohorts consisted of 702 participants: 351 ASD and 351 controls. Image-based classification was performed using one-channel and two-channel inputs to 3D-DenseNet deep learning networks. The models were trained and tested using tenfold cross-validation. Two-channel models were twinned with combinations of structural MRI (sMRI) maps and amplitude of low-frequency fluctuations (ALFF) or fractional ALFF (fALFF) maps from resting-state functional MRI (rs-fMRI). All models produced classification accuracy that exceeded 65.1%. The two-channel ALFF-sMRI model achieved the highest mean accuracy of 76.9% ± 2.34. The one-channel ALFF-based model alone had mean accuracy of 72% ± 3.1. This study leveraged the ABIDE dataset to produce ASD classification results that are comparable and/or exceed literature values. The deep learning approach was conducive to diverse neuroimaging inputs. Findings reveal that the ALFF-sMRI two-channel model outperformed all others.
- Open Access
- 3 May 2024
- Published by Frontiers Media SA in Frontiers in Systems Neuroscience
- Vol. 18, 1327674
A hybrid boundary element-finite element approach for solving the EEG forward problem in brain modeling
This article introduces a hybrid BE-FE method for solving the EEG forward problem, leveraging the strengths of both the Boundary Element Method (BEM) and Finite Element Method (FEM). FEM accurately models complex and anisotropic tissue properties for realistic head geometries, while BEM excels in handling isotropic tissue regions and dipolar sources efficiently. The proposed hybrid method divides regions into homogeneous boundary element (BE) regions that include sources and heterogeneous anisotropic finite element (FE) regions. So, BEM models the brain, including dipole sources, and FEM models other head layers. Validation includes inhomogeneous isotropic/anisotropic three- and four-layer spherical head models, and a four-layer MRI-based realistic head model. Results for six dipole eccentricities and two orientations are computed using BEM, FEM, and hybrid BE-FE method. Statistical analysis, comparing error criteria of RDM and MAG, reveals notable improvements using the hybrid FE-BE method. In the spherical head model, the hybrid BE-FE method compared with FEM demonstrates enhancements of at least 1.05 and 38.31% in RDM and MAG criteria, respectively. Notably, in the anisotropic four-layer head model, improvements reach a maximum of 88.3% for RDM and 93.27% for MAG over FEM. Moreover, in the anisotropic four-layer realistic head model, the proposed hybrid method exhibits 55.4% improvement in RDM and 89.3% improvement in MAG compared to FEM. These findings underscore the proposed method is a promising approach for solving the realistic EEG forward problems, advancing neuroimaging techniques and enhancing understanding of brain function.
- 3 May 2024
- Published by Frontiers Media SA in Frontiers in Systems Neuroscience
- Vol. 18, 1327674
Open AccessA hybrid boundary element-finite element approach for solving the EEG forward problem in brain modeling
This article introduces a hybrid BE-FE method for solving the EEG forward problem, leveraging the strengths of both the Boundary Element Method (BEM) and Finite Element Method (FEM). FEM accurately models complex and anisotropic tissue properties for realistic head geometries, while BEM excels in handling isotropic tissue regions and dipolar sources efficiently. The proposed hybrid method divides regions into homogeneous boundary element (BE) regions that include sources and heterogeneous anisotropic finite element (FE) regions. So, BEM models the brain, including dipole sources, and FEM models other head layers. Validation includes inhomogeneous isotropic/anisotropic three- and four-layer spherical head models, and a four-layer MRI-based realistic head model. Results for six dipole eccentricities and two orientations are computed using BEM, FEM, and hybrid BE-FE method. Statistical analysis, comparing error criteria of RDM and MAG, reveals notable improvements using the hybrid FE-BE method. In the spherical head model, the hybrid BE-FE method compared with FEM demonstrates enhancements of at least 1.05 and 38.31% in RDM and MAG criteria, respectively. Notably, in the anisotropic four-layer head model, improvements reach a maximum of 88.3% for RDM and 93.27% for MAG over FEM. Moreover, in the anisotropic four-layer realistic head model, the proposed hybrid method exhibits 55.4% improvement in RDM and 89.3% improvement in MAG compared to FEM. These findings underscore the proposed method is a promising approach for solving the realistic EEG forward problems, advancing neuroimaging techniques and enhancing understanding of brain function.
- 21 February 2024
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 92
A novel stretching factor estimator based on an adaptive bisection method for ultrasound strain imaging
- 21 February 2024
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 92
A novel stretching factor estimator based on an adaptive bisection method for ultrasound strain imaging