Lailatul Qadri Zakaria
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lailatul Qadri Zakaria has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Melinda Melinda
Department of Electrical and Computer Engineering, Universitas Syiah Kuala, Banda Aceh, Indonesia
3 shared publications - Qiong Hong
Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia (UKM), Bangi 43600, Malaysia
3 shared publications Sabrina Tiun
3 shared publications- Arya Suyanda
Department of Electrical and Computer Engineering, Universitas Syiah Kuala, Banda Aceh, Indonesia, Indonesia
2 shared publications Ridha Hussein Chassab
Univ Kebangsaan Malaysia, Fac Informat Sci & Technol, Bangi, Selangor, Malaysia
2 shared publications- Sabrina Tiun
Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia (UKM), Bangi 43600, Malaysia
2 shared publications
Latest Publications
- 13 October 2026
- Published byĀ SAGE PublicationsĀ inĀ Applied Ontology
An Ontology for Rehabilitation Assessment: Case Study Palestine
Physiotherapists need appropriate data management for patient management with disabilities. Rehabilitation principles, such as evaluation, goal setting, treatment, and results, are the framework for therapy development. Nevertheless, the suitability of controlled rehabilitation studies for all instances of real-life procedures has been an issue that still requires attention because of the differences in the disease conditions that patients may have. In Palestine, such rehabilitation data is often very unstructured, semi-structured, and manually logged, which makes integration within the patient's medical report a challenging task. The existing rehabilitation ontologies, in the form of physical medicine and rehabilitation, do not completely support standard physical therapy (PT) and occupational therapy (OT) assessments as used in Palestine. This is because their combination with the ontology was not applied. To address this problem, a new rehabilitation assessment ontology (RAO) system is suggested as the foundation, which will be composed of PT and OT standard assessments alongside vital patient information. The RAO was developed using the 101-ontology method by conducting interviews across the three rehabilitation centers in Palestine: Specialized Al-Ibtisameh Al-Jamelah Physiotherapy Center (BJC), Hashmi for Specialized Physiotherapy and Spinal Manipulation (HSPT), and Future Magic. The knowledge was collected from a patient profile containing a total of 120 and was utilized to create automatic ontology population rehab gazetteers. The RAO was assessed by a group of medical and ontology specialists, who explained that it was completed along the dimensions of accuracy, extensibility, usability, and completeness. This small scale accurately answered 36 well-defined competency questions. Among others, the RAO solution system presents a structured, interoperable, and improved rehabilitation data organization, simplifying queries to bring about better patient care in the three centers.
- 13 October 2026
- Published byĀ SAGE PublicationsĀ inĀ Applied Ontology
An Ontology for Rehabilitation Assessment: Case Study Palestine
Physiotherapists need appropriate data management for patient management with disabilities. Rehabilitation principles, such as evaluation, goal setting, treatment, and results, are the framework for therapy development. Nevertheless, the suitability of controlled rehabilitation studies for all instances of real-life procedures has been an issue that still requires attention because of the differences in the disease conditions that patients may have. In Palestine, such rehabilitation data is often very unstructured, semi-structured, and manually logged, which makes integration within the patient's medical report a challenging task. The existing rehabilitation ontologies, in the form of physical medicine and rehabilitation, do not completely support standard physical therapy (PT) and occupational therapy (OT) assessments as used in Palestine. This is because their combination with the ontology was not applied. To address this problem, a new rehabilitation assessment ontology (RAO) system is suggested as the foundation, which will be composed of PT and OT standard assessments alongside vital patient information. The RAO was developed using the 101-ontology method by conducting interviews across the three rehabilitation centers in Palestine: Specialized Al-Ibtisameh Al-Jamelah Physiotherapy Center (BJC), Hashmi for Specialized Physiotherapy and Spinal Manipulation (HSPT), and Future Magic. The knowledge was collected from a patient profile containing a total of 120 and was utilized to create automatic ontology population rehab gazetteers. The RAO was assessed by a group of medical and ontology specialists, who explained that it was completed along the dimensions of accuracy, extensibility, usability, and completeness. This small scale accurately answered 36 well-defined competency questions. Among others, the RAO solution system presents a structured, interoperable, and improved rehabilitation data organization, simplifying queries to bring about better patient care in the three centers.
- 8 August 2026
- Published byĀ Springer Nature
- p. 25-38
Sarcasm Markers for Malay Social Media Text: Establishing a Classification Baseline
- 8 August 2026
- Published byĀ Springer Nature
- p. 25-38
Sarcasm Markers for Malay Social Media Text: Establishing a Classification Baseline
- Open Access
- 31 July 2026
- Published byĀ PeerJĀ inĀ PeerJ Computer Science
- Vol.Ā 12, e3935
A systematic review of multimodal depression detection: from technological pipelines to the clinical depression management cycle
The effective management of depressionāa high-prevalence mental disorder defined by persistent low mood and complex psychosomatic symptomsāposes a significant global challenge. It is necessary to conduct objective, multi-source assessments throughout the entire depression management cycle (DMC), from early screening to long-term prognosis. Multimodal depression detection (MMDD) has emerged as a critical enabling technology that leverages artificial intelligence (AI) to integrate these multi-source data streams. We conducted a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-compliant search across major digital libraries (2015ā2025; through Oct 23, 2025) and included 162 studies, synthesizing the technical pipelineādatasets, feature engineering, models, and fusion strategies. To unify fragmented perspectives, we introduce the AIādriven depression management cycle (AI-DMC), which maps MMDD applications onto two task paradigms: low-fidelity, large-scale monitoring (screening, prognosis) and high-fidelity, high-precision detection (diagnosis support, treatment assessment). Grounded in AI-DMC, we outline a clinically oriented roadmap and identify system-level gapsādata and annotation ecosystems, trustworthy fusion and interpretability, and translation from monitoring to clinical actionāproviding forward-looking guidance for research and deployment.
- 31 July 2026
- Published byĀ PeerJĀ inĀ PeerJ Computer Science
- Vol.Ā 12, e3935
Open AccessA systematic review of multimodal depression detection: from technological pipelines to the clinical depression management cycle
The effective management of depressionāa high-prevalence mental disorder defined by persistent low mood and complex psychosomatic symptomsāposes a significant global challenge. It is necessary to conduct objective, multi-source assessments throughout the entire depression management cycle (DMC), from early screening to long-term prognosis. Multimodal depression detection (MMDD) has emerged as a critical enabling technology that leverages artificial intelligence (AI) to integrate these multi-source data streams. We conducted a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-compliant search across major digital libraries (2015ā2025; through Oct 23, 2025) and included 162 studies, synthesizing the technical pipelineādatasets, feature engineering, models, and fusion strategies. To unify fragmented perspectives, we introduce the AIādriven depression management cycle (AI-DMC), which maps MMDD applications onto two task paradigms: low-fidelity, large-scale monitoring (screening, prognosis) and high-fidelity, high-precision detection (diagnosis support, treatment assessment). Grounded in AI-DMC, we outline a clinically oriented roadmap and identify system-level gapsādata and annotation ecosystems, trustworthy fusion and interpretability, and translation from monitoring to clinical actionāproviding forward-looking guidance for research and deployment.
- Open Access
- 22 May 2026
- Published byĀ MDPI AGĀ inĀ Information
- Vol.Ā 17 Ā (6) , 516
NS-Dep-KAN: An Explainable Neuro-Symbolic Framework with KolmogorovāArnold Networks for DSM-Guided Depression Assessment
Automated depression assessment is critical for scalable mental healthcare but faces dual challenges: the lack of clinical interpretability in āblack-boxā deep learning models and the excessive computational cost of large-scale fusion architectures. To bridge this gap, we propose NS-Dep-KAN, a novel neuro-symbolic framework that harmonizes DSM-5-guided reasoning with KolmogorovāArnold Networks (KANs). Our approach leverages a Large Language Model (LLM) to extract symbolic symptom evidence aligned with diagnostic criteria, which then guides the aggregation of multimodal features from frozen pretrained encoders (WavLM and Qwen). Unlike traditional Multi-Layer Perceptrons, the proposed KAN prediction head employs learnable B-spline activation functions to capture complex nonlinear symptomāseverity mappings with extreme parameter efficiency. Evaluations on the DAIC-WOZ benchmark demonstrate that NS-Dep-KAN achieves state-of-the-art performance among audio-text models (MAE 2.69, 13.5% improvement over the three-modality baseline MSGAF at MAE 3.11), with only ā¼4.9 K trainable parameters. Moreover, the framework offers inherent interpretability, revealing granular symptom contribution profiles that align with clinical intuition. This work establishes a path toward explainable trustworthy AI for mental health screening.
- 22 May 2026
- Published byĀ MDPI AGĀ inĀ Information
- Vol.Ā 17 Ā (6) , 516
Open AccessNS-Dep-KAN: An Explainable Neuro-Symbolic Framework with KolmogorovāArnold Networks for DSM-Guided Depression Assessment
Automated depression assessment is critical for scalable mental healthcare but faces dual challenges: the lack of clinical interpretability in āblack-boxā deep learning models and the excessive computational cost of large-scale fusion architectures. To bridge this gap, we propose NS-Dep-KAN, a novel neuro-symbolic framework that harmonizes DSM-5-guided reasoning with KolmogorovāArnold Networks (KANs). Our approach leverages a Large Language Model (LLM) to extract symbolic symptom evidence aligned with diagnostic criteria, which then guides the aggregation of multimodal features from frozen pretrained encoders (WavLM and Qwen). Unlike traditional Multi-Layer Perceptrons, the proposed KAN prediction head employs learnable B-spline activation functions to capture complex nonlinear symptomāseverity mappings with extreme parameter efficiency. Evaluations on the DAIC-WOZ benchmark demonstrate that NS-Dep-KAN achieves state-of-the-art performance among audio-text models (MAE 2.69, 13.5% improvement over the three-modality baseline MSGAF at MAE 3.11), with only ā¼4.9 K trainable parameters. Moreover, the framework offers inherent interpretability, revealing granular symptom contribution profiles that align with clinical intuition. This work establishes a path toward explainable trustworthy AI for mental health screening.
- Open Access
- 18 May 2026
- Published byĀ MDPI AGĀ inĀ Applied Sciences
- Vol.Ā 16 Ā (10) , 5018
A Systematic Review of Transformer-Based Models for Depression Detection
Depression is a critical global public health challenge, and the demand for accurate automated detection methods has generated considerable research interest in Transformer-based models. Despite their substantial promise, a comprehensive investigation into their architectural efficacy, intrinsic mechanisms, and barriers to practical implementation remains lacking. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, this systematic review was conducted across six databases (IEEE Xplore, Elsevier, Springer, MDPI, PubMed, and arXiv). The final search was performed in October 2025, covering English-language empirical studies published between 2020 and 2025 that employed Transformer-based architectures for depression detection. Risk of bias and methodological quality were independently appraised by two authors using a six-dimension structured rubric, with disagreements resolved by a third author. Findings were narratively synthesized given substantial cross-study heterogeneity. This systematic review analyzed 46 studies and provided the first comprehensive, mechanism-level, architecturally stratified comparison of encoder-only, decoder-only, hybrid, and multimodal fusion paradigms, examining self-attention dynamics and transfer learning strategies. Since 2019, these frameworks have evolved from text-centric approaches to advanced multimodal systems. Encoder-only models show consistently strong results in high-throughput text-based screening, decoder-only models demonstrate stronger few-shot learning capabilities, hybrid architectures show the highest observed median performance in clinical interview settings across the reviewed studies, and multimodal fusion systems offer complementary advantages when heterogeneous signal integration is critical. These trends are task-contextualized and should not be interpreted as unconditional rankings, given heterogeneity in evaluation metrics and tasks across studies. Nonetheless, four principal challenges hinder clinical translation: overreliance on self-reported data, cross-linguistic bias, absence of uncertainty quantification, and substantial computational overhead. Future efforts should shift from incremental benchmark improvements toward clinical utility through standardized psychiatric validation, uncertainty-aware architectures, fairness-enforced training across diverse populations, and the integration of Transformer-based models with wearable and mobile health data to improve detection stability and reduce translational risk. This systematic review was registered on the Open Science Framework (OSF; DOI: 10.17605/OSF.IO/SYF9N). This research was funded by the Faculty of Information Science and Technology and by Universiti Kebangsaan Malaysia under Grant TAP-K014364.
- 18 May 2026
- Published byĀ MDPI AGĀ inĀ Applied Sciences
- Vol.Ā 16 Ā (10) , 5018
Open AccessA Systematic Review of Transformer-Based Models for Depression Detection
Depression is a critical global public health challenge, and the demand for accurate automated detection methods has generated considerable research interest in Transformer-based models. Despite their substantial promise, a comprehensive investigation into their architectural efficacy, intrinsic mechanisms, and barriers to practical implementation remains lacking. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, this systematic review was conducted across six databases (IEEE Xplore, Elsevier, Springer, MDPI, PubMed, and arXiv). The final search was performed in October 2025, covering English-language empirical studies published between 2020 and 2025 that employed Transformer-based architectures for depression detection. Risk of bias and methodological quality were independently appraised by two authors using a six-dimension structured rubric, with disagreements resolved by a third author. Findings were narratively synthesized given substantial cross-study heterogeneity. This systematic review analyzed 46 studies and provided the first comprehensive, mechanism-level, architecturally stratified comparison of encoder-only, decoder-only, hybrid, and multimodal fusion paradigms, examining self-attention dynamics and transfer learning strategies. Since 2019, these frameworks have evolved from text-centric approaches to advanced multimodal systems. Encoder-only models show consistently strong results in high-throughput text-based screening, decoder-only models demonstrate stronger few-shot learning capabilities, hybrid architectures show the highest observed median performance in clinical interview settings across the reviewed studies, and multimodal fusion systems offer complementary advantages when heterogeneous signal integration is critical. These trends are task-contextualized and should not be interpreted as unconditional rankings, given heterogeneity in evaluation metrics and tasks across studies. Nonetheless, four principal challenges hinder clinical translation: overreliance on self-reported data, cross-linguistic bias, absence of uncertainty quantification, and substantial computational overhead. Future efforts should shift from incremental benchmark improvements toward clinical utility through standardized psychiatric validation, uncertainty-aware architectures, fairness-enforced training across diverse populations, and the integration of Transformer-based models with wearable and mobile health data to improve detection stability and reduce translational risk. This systematic review was registered on the Open Science Framework (OSF; DOI: 10.17605/OSF.IO/SYF9N). This research was funded by the Faculty of Information Science and Technology and by Universiti Kebangsaan Malaysia under Grant TAP-K014364.
- Open Access
- 10 April 2026
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Jurnal Teknokes
- Vol.Ā 19 Ā (2) , 061-071
Classification of Autism Spectrum Disorder (ASD) in Children Using the VGG19 CNN Model Based on Facial Landmarks of the Eye and Forehead Areas
Early detection of Autism Spectrum Disorder (ASD) is a crucial challenge in child development interventions because conventional screening methods are often subjective and prone to assessor bias. This study proposes an objective solution in the form of a deep learning approach for automatic ASD classification using facial landmark representations that focus exclusively on the eye and forehead areas. The selection of these areas is based on the eye avoidance hypothesis, which states that these regions contain very rich diagnostic information and behavioral biomarkers related to the ASD phenotype. The pre-processing stage involves isolating the eye and forehead areas using Dlib 68-landmark detection to eliminate background visual noise, followed by detailed topological visualization using MediaPipe Face Mesh with 478 landmark points as the model input. The Convolutional Neural Network (CNN) architecture used is the VGG19 model modified with transfer learning techniques and the addition of Dropout layers to improve efficiency and prevent overfitting. The model was trained on a primary dataset of 1,238 images collected under controlled conditions from children in Banda Aceh. The test results showed very promising performance with an overall accuracy of 94.35%. Specifically, the model achieved a recall (sensitivity) of 95.24%, a precision of 93.75%, and an AUC score of 0.9831. This high sensitivity is crucial in a medical context to minimize the risk of misdetection of positive cases. These results demonstrate that landmark visualization in the eye and forehead areas with the VGG19 model is a highly effective, accurate, and practical method for serving as an economical early screening tool for ASD.
- 10 April 2026
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Jurnal Teknokes
- Vol.Ā 19 Ā (2) , 061-071
Open AccessClassification of Autism Spectrum Disorder (ASD) in Children Using the VGG19 CNN Model Based on Facial Landmarks of the Eye and Forehead Areas
Early detection of Autism Spectrum Disorder (ASD) is a crucial challenge in child development interventions because conventional screening methods are often subjective and prone to assessor bias. This study proposes an objective solution in the form of a deep learning approach for automatic ASD classification using facial landmark representations that focus exclusively on the eye and forehead areas. The selection of these areas is based on the eye avoidance hypothesis, which states that these regions contain very rich diagnostic information and behavioral biomarkers related to the ASD phenotype. The pre-processing stage involves isolating the eye and forehead areas using Dlib 68-landmark detection to eliminate background visual noise, followed by detailed topological visualization using MediaPipe Face Mesh with 478 landmark points as the model input. The Convolutional Neural Network (CNN) architecture used is the VGG19 model modified with transfer learning techniques and the addition of Dropout layers to improve efficiency and prevent overfitting. The model was trained on a primary dataset of 1,238 images collected under controlled conditions from children in Banda Aceh. The test results showed very promising performance with an overall accuracy of 94.35%. Specifically, the model achieved a recall (sensitivity) of 95.24%, a precision of 93.75%, and an AUC score of 0.9831. This high sensitivity is crucial in a medical context to minimize the risk of misdetection of positive cases. These results demonstrate that landmark visualization in the eye and forehead areas with the VGG19 model is a highly effective, accurate, and practical method for serving as an economical early screening tool for ASD.
- Open Access
- 1 August 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol.Ā 7 Ā (4) , 1045-1059
Multispectral Classification based on H20 and H20 with NaOH Using Image Segmentation and Ensemble Learning EfficientNetV2, Resnet50, MobileNetV3
High Multispectral imaging has become a promising approach in liquid classification, particularly in distinguishing visually similar but subtly spectrally distinct solutions, such as pure water (HāO) and water mixed with sodium hydroxide (HāO with NaOH). This study proposed a classification system based on image segmentation and deep learning, utilizing three leading Convolutional Neural Network (CNN) architectures: ResNet 50, EfficientNetV2, and MobileNetV3. Before classification, each multispectral image was processed through color segmentation in HSV space to highlight the dominant spectral, especially in the hue range of 110 170. The model was trained using a data augmentation scheme and optimized with the Adam algorithm, a batch size of 32, and a sigmoid activation function. The dataset consists of 807 images, including 295 HāO images and 512 HāO with NaOH images, which were divided into training (64%), validation (16%), and testing (20%) data. Experimental results show that ResNet50 achieves the highest performance, with an accuracy of 93.83% and an F1 score of 93.67%, particularly in identifying alkaline pollution. EfficientNetV2 achieved the lowest loss (0.2001) and exhibited balanced performance across classes, while MobileNetV3, despite being a lightweight model, remained competitive with a recall of 0.97 in the HāO with NaOH class. Further evaluation with Grad CAM reveals that all models focus on the most critical spectral areas of the segmentation results. These findings support the effectiveness of combining color-based segmentation and CNN in the spectral classification of liquids. This research is expected to serve as a stepping stone in the development of an efficient and accurate automatic liquid classification system for both laboratory and industrial applications.
- 1 August 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol.Ā 7 Ā (4) , 1045-1059
Open AccessMultispectral Classification based on H20 and H20 with NaOH Using Image Segmentation and Ensemble Learning EfficientNetV2, Resnet50, MobileNetV3
High Multispectral imaging has become a promising approach in liquid classification, particularly in distinguishing visually similar but subtly spectrally distinct solutions, such as pure water (HāO) and water mixed with sodium hydroxide (HāO with NaOH). This study proposed a classification system based on image segmentation and deep learning, utilizing three leading Convolutional Neural Network (CNN) architectures: ResNet 50, EfficientNetV2, and MobileNetV3. Before classification, each multispectral image was processed through color segmentation in HSV space to highlight the dominant spectral, especially in the hue range of 110 170. The model was trained using a data augmentation scheme and optimized with the Adam algorithm, a batch size of 32, and a sigmoid activation function. The dataset consists of 807 images, including 295 HāO images and 512 HāO with NaOH images, which were divided into training (64%), validation (16%), and testing (20%) data. Experimental results show that ResNet50 achieves the highest performance, with an accuracy of 93.83% and an F1 score of 93.67%, particularly in identifying alkaline pollution. EfficientNetV2 achieved the lowest loss (0.2001) and exhibited balanced performance across classes, while MobileNetV3, despite being a lightweight model, remained competitive with a recall of 0.97 in the HāO with NaOH class. Further evaluation with Grad CAM reveals that all models focus on the most critical spectral areas of the segmentation results. These findings support the effectiveness of combining color-based segmentation and CNN in the spectral classification of liquids. This research is expected to serve as a stepping stone in the development of an efficient and accurate automatic liquid classification system for both laboratory and industrial applications.
- Open Access
- 3 May 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol.Ā 7 Ā (3) , 925-939
EEG Performance Signal Analysis for Diagnosing Autism Spectrum Disorder using Butterworth and Empirical Mode Decomposition
Electroencephalography (EEG) is a technique used to measure electrical activity in the brain by placing electrodes on the scalp. EEG plays an essential role in analyzing a variety of neurological conditions, including autism spectrum disorder (ASD). However, in the recording process, EEG signals are often contaminated by noise, hindering further analysis. Therefore, an effective signal processing method is needed to improve the data quality before feature extraction is performed. This study applied the Butterworth Band-Pass Filter (BPF) as a preprocessing method to reduce noise in EEG signals and then used the Empirical Mode Decomposition (EMD) method to extract relevant features. The performance of this method was evaluated using three main parameters, namely Mean Square Error (MSE), Mean Absolute Error (MAE), and Signal-to-Noise Ratio (SNR). The results showed that EMD was able to retain important information in EEG signals better than signals that only passed through the BPF filtration stage. EMD produces lower MAE and MSE values than Butterworth, suggesting that this method is more accurate in maintaining the original shape of the signal. In subject 3, EMD recorded the lowest MAE of 0.622 compared to Butterworth, which reached 20.0, and the MSE value of 0.655 compared to 771.5 for Butterworth. In addition, EMD also produced a higher SNR, with the highest value of 23,208 in subject 5, compared to Butterworth, which reached only 1,568. These results prove that the combination of BPF as a preprocessing method and EMD as a feature extraction method is more effective in maintaining EEG signal quality and improving analysis accuracy compared to the use of the Butterworth Band-Pass Filter alone.
- 3 May 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol.Ā 7 Ā (3) , 925-939
Open AccessEEG Performance Signal Analysis for Diagnosing Autism Spectrum Disorder using Butterworth and Empirical Mode Decomposition
Electroencephalography (EEG) is a technique used to measure electrical activity in the brain by placing electrodes on the scalp. EEG plays an essential role in analyzing a variety of neurological conditions, including autism spectrum disorder (ASD). However, in the recording process, EEG signals are often contaminated by noise, hindering further analysis. Therefore, an effective signal processing method is needed to improve the data quality before feature extraction is performed. This study applied the Butterworth Band-Pass Filter (BPF) as a preprocessing method to reduce noise in EEG signals and then used the Empirical Mode Decomposition (EMD) method to extract relevant features. The performance of this method was evaluated using three main parameters, namely Mean Square Error (MSE), Mean Absolute Error (MAE), and Signal-to-Noise Ratio (SNR). The results showed that EMD was able to retain important information in EEG signals better than signals that only passed through the BPF filtration stage. EMD produces lower MAE and MSE values than Butterworth, suggesting that this method is more accurate in maintaining the original shape of the signal. In subject 3, EMD recorded the lowest MAE of 0.622 compared to Butterworth, which reached 20.0, and the MSE value of 0.655 compared to 771.5 for Butterworth. In addition, EMD also produced a higher SNR, with the highest value of 23,208 in subject 5, compared to Butterworth, which reached only 1,568. These results prove that the combination of BPF as a preprocessing method and EMD as a feature extraction method is more effective in maintaining EEG signal quality and improving analysis accuracy compared to the use of the Butterworth Band-Pass Filter alone.
- Open Access
- 24 February 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol.Ā 7 Ā (1) , 102-112
Implementation of Vision Transformer for Early Detection of Autism Based on EEG Signal Heatmap Visualization
Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder characterized by difficulties in social interaction, communication, and repetitive behavioral patterns. Early detection of ASD is crucial for improving the quality of life of affected individuals and alleviating the burden on their families. This study proposes a computer-aided diagnostic system for ASD by applying a pre-trained Vision Transformer (ViT-B/16) architecture to EEG signal data obtained from King Abdul Aziz University. The dataset comprises EEG recordings from 16 subjects (8 normal and 8 ASD) that have undergone preprocessingāincluding filtering using the Discrete Wavelet Transform (DWT), segmentation (windowing), and conversion into heatmap representationsāand were subsequently partitioned into training, validation, and testing subsets. The ViT model was trained for 100 epochs with a batch size of 16, using the AdamW optimizer and the CrossEntropy loss function, while two learning rate configurations (0.0001 and 0.00001) were evaluated; the best-performing weights were selected based on the lowest validation loss. Test results indicate that the model trained with a learning rate of 0.00001 achieved a testing accuracy of 99.53%, accompanied by excellent precision, specificity, recall, and f1-score, thereby demonstrating strong generalization capabilities and minimal overfitting. Future research is recommended to incorporate locally sourced datasets and to further customize the ViT architecture through comprehensive hyperparameter tuning, with the aim of developing a mobile application to support clinical ASD diagnosis.
- 24 February 2025
- Published byĀ Poltekkes Kemenkes SurabayaĀ inĀ Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol.Ā 7 Ā (1) , 102-112
Open AccessImplementation of Vision Transformer for Early Detection of Autism Based on EEG Signal Heatmap Visualization
Autism Spectrum Disorder (ASD) is a complex neurodevelopmental disorder characterized by difficulties in social interaction, communication, and repetitive behavioral patterns. Early detection of ASD is crucial for improving the quality of life of affected individuals and alleviating the burden on their families. This study proposes a computer-aided diagnostic system for ASD by applying a pre-trained Vision Transformer (ViT-B/16) architecture to EEG signal data obtained from King Abdul Aziz University. The dataset comprises EEG recordings from 16 subjects (8 normal and 8 ASD) that have undergone preprocessingāincluding filtering using the Discrete Wavelet Transform (DWT), segmentation (windowing), and conversion into heatmap representationsāand were subsequently partitioned into training, validation, and testing subsets. The ViT model was trained for 100 epochs with a batch size of 16, using the AdamW optimizer and the CrossEntropy loss function, while two learning rate configurations (0.0001 and 0.00001) were evaluated; the best-performing weights were selected based on the lowest validation loss. Test results indicate that the model trained with a learning rate of 0.00001 achieved a testing accuracy of 99.53%, accompanied by excellent precision, specificity, recall, and f1-score, thereby demonstrating strong generalization capabilities and minimal overfitting. Future research is recommended to incorporate locally sourced datasets and to further customize the ViT architecture through comprehensive hyperparameter tuning, with the aim of developing a mobile application to support clinical ASD diagnosis.
- Open Access
- 30 October 2024
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Access
- Vol.Ā 12, 161203-161212
Integrating Bert With CNN and BiLSTM for Explainable Detection of Depression in Social Media Contents
Depression is a prevalent mental health condition that significantly impacts individualsā lives. Early detection of depression is crucial for timely intervention and improved outcomes. However, traditional machine learning approaches are constrained by the limited amount of annotated data and lack of model transparency. This study aims to address these challenges by leveraging social media data and advanced natural language processing techniques to develop effective and explainable models for depression detection. The study focuses on two main objectives. The first objective is to develop and evaluate fine-tuned BERT, BERT-BiLSTM, and BERT-CNN models, and compare their performance with MentalBERT, a state-of-the-art model for mental health detection. The second objective is to observe the key features used by the BERT models to make the decision-making using Transformer Interpretability Beyond Attention Visualization and Average Attention Weight methods. The study utilizes three publicly available datasets: the Depression Reddit Dataset, the Sentiment Analysis for Tweets Dataset, and the Mental Health Corpus. The results demonstrate that the proposed models, especially BERT-BiLSTM and BERT-CNN, achieve superior performance compared to MentalBERT, particularly regarding accuracy and F1-score. The attention map visualizations provide valuable insights into the language patterns and key features associated with depression in social media posts. This study contributes to the mental health field by presenting novel and explainable models for depression detection using social media data. The proposed approaches have the potential to assist mental health professionals in early identification and intervention, ultimately improving the lives of individuals affected by depression.
- 30 October 2024
- Published byĀ Institute of Electrical and Electronics Engineers (IEEE)Ā inĀ IEEE Access
- Vol.Ā 12, 161203-161212
Open AccessIntegrating Bert With CNN and BiLSTM for Explainable Detection of Depression in Social Media Contents
Depression is a prevalent mental health condition that significantly impacts individualsā lives. Early detection of depression is crucial for timely intervention and improved outcomes. However, traditional machine learning approaches are constrained by the limited amount of annotated data and lack of model transparency. This study aims to address these challenges by leveraging social media data and advanced natural language processing techniques to develop effective and explainable models for depression detection. The study focuses on two main objectives. The first objective is to develop and evaluate fine-tuned BERT, BERT-BiLSTM, and BERT-CNN models, and compare their performance with MentalBERT, a state-of-the-art model for mental health detection. The second objective is to observe the key features used by the BERT models to make the decision-making using Transformer Interpretability Beyond Attention Visualization and Average Attention Weight methods. The study utilizes three publicly available datasets: the Depression Reddit Dataset, the Sentiment Analysis for Tweets Dataset, and the Mental Health Corpus. The results demonstrate that the proposed models, especially BERT-BiLSTM and BERT-CNN, achieve superior performance compared to MentalBERT, particularly regarding accuracy and F1-score. The attention map visualizations provide valuable insights into the language patterns and key features associated with depression in social media posts. This study contributes to the mental health field by presenting novel and explainable models for depression detection using social media data. The proposed approaches have the potential to assist mental health professionals in early identification and intervention, ultimately improving the lives of individuals affected by depression.