Ozlem Karabiber Cura
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ozlem Karabiber Cura has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Aydin Akan
Department of Electrical and Electronics Engineering, Izmir University of Economics, 35330 Izmir, Türkiye
7 shared publications Sibel Kocaaslan Atli
2 shared publications- Abdurrahman Gumus
Department of Electrical and Electronics Engineering, Izmir Institute of Technology, Urla, Turkey
1 shared publication - Adnan Yamanoğlu
İZMİR KATİP ÇELEBİ ÜNİVERSİTESİ, TIP FAKÜLTESİ
1 shared publication - Ahmet Kayali
İZMİR KATİP ÇELEBİ ÜNİVERSİTESİ, TIP FAKÜLTESİ
1 shared publication Aydin Akan
1 shared publication
Latest Publications
- 22 September 2026
- Published by Izmir UOD
- p. 128-136
Automated Detection and Size Classification of Pneumothorax in Chest X-ray Images Using a Two-Stage Deep Learning-Based Clinical Decision Support System
Pneumothorax is a life-threatening condition requiring rapid and accurate diagnosis. This study proposes a two-stage deep learning-based clinical decision support system for automated pneumothorax detection and size classification from posteroanterior chest X-ray images, with separate validation on right and left lung datasets. The dataset comprised 1,170 images: 700 healthy and 470 pneumothorax cases, equally divided into size ≤20% and >20%. In Stage 1, a ResNet50-based binary classifier distinguished pneumothorax from healthy with an AUC of 0.9684 (left) and 0.9475 (right). In Stage 2, a DenseNet201- based classifier further classified detected pneumothorax cases into size ≤20% vs. >20%. The full pipeline achieved high specificity (left: 96.84% for non-pneumothorax; right: 97.75% for high-grade pneumothorax). The system demonstrates strong potential as an AI-assisted triage tool, with separate performance reporting for each lung side enabling clinical validation.
- 22 September 2026
- Published by Izmir UOD
- p. 128-136
Automated Detection and Size Classification of Pneumothorax in Chest X-ray Images Using a Two-Stage Deep Learning-Based Clinical Decision Support System
Pneumothorax is a life-threatening condition requiring rapid and accurate diagnosis. This study proposes a two-stage deep learning-based clinical decision support system for automated pneumothorax detection and size classification from posteroanterior chest X-ray images, with separate validation on right and left lung datasets. The dataset comprised 1,170 images: 700 healthy and 470 pneumothorax cases, equally divided into size ≤20% and >20%. In Stage 1, a ResNet50-based binary classifier distinguished pneumothorax from healthy with an AUC of 0.9684 (left) and 0.9475 (right). In Stage 2, a DenseNet201- based classifier further classified detected pneumothorax cases into size ≤20% vs. >20%. The full pipeline achieved high specificity (left: 96.84% for non-pneumothorax; right: 97.75% for high-grade pneumothorax). The system demonstrates strong potential as an AI-assisted triage tool, with separate performance reporting for each lung side enabling clinical validation.
- 22 September 2026
- Published by Izmir UOD
- p. 117-127
Quantum Probability Image Encoding for Multivariate Time-Series Classification: Regime Diagnostics Across Five UEA Benchmarks
Quantum Probability Image Encoding (QPIE) was developed for spatial multi-channel data such as RGB medical images. We test whether it transfers without modification to multivariate time-series classification (MTSC): each fixed-stride temporal patch is treated as a pixel, the three accelerometer or pen-force channels replace the three color channels, and the five 3-qubit encoding schemes (Sep, CRyE, GBE, CP-2L, CSE) and multi-basis Z+X+Y measurement pipeline are reused unchanged. Across five 3-channel UEA benchmarks -the complete 3-dimensional UEA subset -the central result is negative. On the four datasets where the encoding is functional, no entangling scheme beats the separable baseline (Sep) at p < 0.05, and the data-conditional scheme (CSE) falls significantly below Sep. The image- domain channel-correlation mechanism that makes CSE useful on RGB data does not operate on temporal sensor signals, whose inter-channel correlations are sensor-physical rather than class- conditional. The pooled-to-CNN accuracy gap splits the datasets into regimes: below 5 points where per-patch statistics carry the label (Epilepsy, CharacterTrajectories), and 26–38 points where temporal ordering does (UWaveGestureLibrary, Handwriting). On EthanolConcentration both rungs collapse to near-random, a third regime in which patch subsampling of the 1751-step series discards the discriminative spectral detail. On absolute accuracy QPIE is competitive in data-scarce settings but reaches no new state of the art. It trails InceptionTime by 5.6 points on the native UWave split and LITEMVTime by 2.9 points on Epilepsy (Ismail-Fawaz et al., 2025); on Handwriting it exceeds every deep-learning baseline in the LITE survey yet stays below the strongest convolutional classifiers (Ruiz et al., 2021). Multivariate ROCKET run on the raw 3-channel series (same splits) matches or beats QPIE on every benchmark except CharacterTrajectories. With 896 training samples -the same gesture instances stacked as three channels (the UWaveGestureLibraryAll benchmark) -the separable CNN ensemble reaches 97.5%, within a point of multivariate ROCKET on the raw 3-channel stack (98.4%) (Dempster et al., 2020).
- 22 September 2026
- Published by Izmir UOD
- p. 117-127
Quantum Probability Image Encoding for Multivariate Time-Series Classification: Regime Diagnostics Across Five UEA Benchmarks
Quantum Probability Image Encoding (QPIE) was developed for spatial multi-channel data such as RGB medical images. We test whether it transfers without modification to multivariate time-series classification (MTSC): each fixed-stride temporal patch is treated as a pixel, the three accelerometer or pen-force channels replace the three color channels, and the five 3-qubit encoding schemes (Sep, CRyE, GBE, CP-2L, CSE) and multi-basis Z+X+Y measurement pipeline are reused unchanged. Across five 3-channel UEA benchmarks -the complete 3-dimensional UEA subset -the central result is negative. On the four datasets where the encoding is functional, no entangling scheme beats the separable baseline (Sep) at p < 0.05, and the data-conditional scheme (CSE) falls significantly below Sep. The image- domain channel-correlation mechanism that makes CSE useful on RGB data does not operate on temporal sensor signals, whose inter-channel correlations are sensor-physical rather than class- conditional. The pooled-to-CNN accuracy gap splits the datasets into regimes: below 5 points where per-patch statistics carry the label (Epilepsy, CharacterTrajectories), and 26–38 points where temporal ordering does (UWaveGestureLibrary, Handwriting). On EthanolConcentration both rungs collapse to near-random, a third regime in which patch subsampling of the 1751-step series discards the discriminative spectral detail. On absolute accuracy QPIE is competitive in data-scarce settings but reaches no new state of the art. It trails InceptionTime by 5.6 points on the native UWave split and LITEMVTime by 2.9 points on Epilepsy (Ismail-Fawaz et al., 2025); on Handwriting it exceeds every deep-learning baseline in the LITE survey yet stays below the strongest convolutional classifiers (Ruiz et al., 2021). Multivariate ROCKET run on the raw 3-channel series (same splits) matches or beats QPIE on every benchmark except CharacterTrajectories. With 896 training samples -the same gesture instances stacked as three channels (the UWaveGestureLibraryAll benchmark) -the separable CNN ensemble reaches 97.5%, within a point of multivariate ROCKET on the raw 3-channel stack (98.4%) (Dempster et al., 2020).
- 22 May 2026
- Published by Erciyes Universitesi (DergiPark) in Erciyes Üniversitesi Fen Bilimleri Enstitüsü Fen Bilimleri Dergisi
- Vol. 42 (2)
Automated Schizophrenia Detection via Resnet50-based Ensemble Learning Using Multi-entropy EEG Feature Maps
Background/Aim: A complicated neuropsychiatric condition, schizophrenia is commonly investigated using one-dimensional EEG signals. In this work, a novel deep learning framework for automated schizophrenia detection is proposed, integrating an ensemble model based on ResNet50 with multi-entropy-based feature maps derived from 10-second EEG segments.Methods: Six entropy measures—approximate, bubble, fuzzy, permutation, sample, and spectral entropy—are computed to create two-dimensional EEG feature maps (EEG-FMs). A distinct ResNet50 model processes each EEG-FM, and soft voting is utilized for final predictions. The framework's effectiveness is evaluated using a leave-one-group-out cross-validation (LOGO-CV) method on two EEG datasets.Results: The ensemble framework exhibits high sensitivity and specificity, achieving average accuracies of 81.4% on Dataset 1 and 95.7% on Dataset 2. It evaluates the framework's performance across varied EEG distributions using a combined dataset and the Wilcoxon signed-rank test. The soft voting generally yields more balanced results than individual entropy-based models, although performance declines with increased dataset variability. An independent third EEG dataset (Dataset 3) is used to further assess the framework under differing conditions.Conclusion: The integration of entropy-based EEG features into the deep learning ensemble improves schizophrenia detection.
- 22 May 2026
- Published by Erciyes Universitesi (DergiPark) in Erciyes Üniversitesi Fen Bilimleri Enstitüsü Fen Bilimleri Dergisi
- Vol. 42 (2)
Automated Schizophrenia Detection via Resnet50-based Ensemble Learning Using Multi-entropy EEG Feature Maps
Background/Aim: A complicated neuropsychiatric condition, schizophrenia is commonly investigated using one-dimensional EEG signals. In this work, a novel deep learning framework for automated schizophrenia detection is proposed, integrating an ensemble model based on ResNet50 with multi-entropy-based feature maps derived from 10-second EEG segments.Methods: Six entropy measures—approximate, bubble, fuzzy, permutation, sample, and spectral entropy—are computed to create two-dimensional EEG feature maps (EEG-FMs). A distinct ResNet50 model processes each EEG-FM, and soft voting is utilized for final predictions. The framework's effectiveness is evaluated using a leave-one-group-out cross-validation (LOGO-CV) method on two EEG datasets.Results: The ensemble framework exhibits high sensitivity and specificity, achieving average accuracies of 81.4% on Dataset 1 and 95.7% on Dataset 2. It evaluates the framework's performance across varied EEG distributions using a combined dataset and the Wilcoxon signed-rank test. The soft voting generally yields more balanced results than individual entropy-based models, although performance declines with increased dataset variability. An independent third EEG dataset (Dataset 3) is used to further assess the framework under differing conditions.Conclusion: The integration of entropy-based EEG features into the deep learning ensemble improves schizophrenia detection.
- 21 April 2026
- Published by Bulent Evcevit University in Karaelmas Science and Engineering Journal
- Vol. 16 (1) , 103-117
Graph Theory Metrics-based Functional Brain Network Analysis in Dyslexia
Developmental dyslexia, a neurodevelopmental disorder affecting 5-10% of school-aged children, involves persistent reading, decoding, and spelling difficulties. Electroencephalography (EEG), especially when combined with graph theory, offers a powerful, non-invasive method for exploring the neural basis of dyslexia. This study examines neuroplastic changes in 14 subjects with dyslexia during a 60-day cognitive training protocol, utilizing EEG-based functional connectivity and brain network analysis. The international 10–20 system is used to capture EEG signals from 14 scalp locations that encompass the frontal, temporal, parietal, and occipital lobes. The EEG recordings from Days 1 and 60 are used to extract a variety of connectivity metrics, including correlation coefficients, covariances, cross-power spectral densities, magnitude-squared coherence, and entropy-based metrics. Additionally, seven graph theory metrics (assortativity, transitivity, clustering coefficient, global efficiency, local efficiency, modularity, and characteristic path length) are calculated using connectivity measures to assess changes in large-scale brain networks. In order to evaluate the impact of cognitive training, EEG-derived features are classified using machine learning models such as logistic regression, decision trees, bagging trees, k-nearest neighbor, and support vector machines. By Day 60, graph-based connectivity tests revealed enhanced long-range functional connectivity and better brain network integration. These results imply that children with dyslexia may have quantifiable neuroplasticity as a result of cognitive instruction. EEG-based graph metrics are also highlighted in the paper as possible indicators for tracking dyslexic children’s cognitive development and the effectiveness of their treatment.
- 21 April 2026
- Published by Bulent Evcevit University in Karaelmas Science and Engineering Journal
- Vol. 16 (1) , 103-117
Graph Theory Metrics-based Functional Brain Network Analysis in Dyslexia
Developmental dyslexia, a neurodevelopmental disorder affecting 5-10% of school-aged children, involves persistent reading, decoding, and spelling difficulties. Electroencephalography (EEG), especially when combined with graph theory, offers a powerful, non-invasive method for exploring the neural basis of dyslexia. This study examines neuroplastic changes in 14 subjects with dyslexia during a 60-day cognitive training protocol, utilizing EEG-based functional connectivity and brain network analysis. The international 10–20 system is used to capture EEG signals from 14 scalp locations that encompass the frontal, temporal, parietal, and occipital lobes. The EEG recordings from Days 1 and 60 are used to extract a variety of connectivity metrics, including correlation coefficients, covariances, cross-power spectral densities, magnitude-squared coherence, and entropy-based metrics. Additionally, seven graph theory metrics (assortativity, transitivity, clustering coefficient, global efficiency, local efficiency, modularity, and characteristic path length) are calculated using connectivity measures to assess changes in large-scale brain networks. In order to evaluate the impact of cognitive training, EEG-derived features are classified using machine learning models such as logistic regression, decision trees, bagging trees, k-nearest neighbor, and support vector machines. By Day 60, graph-based connectivity tests revealed enhanced long-range functional connectivity and better brain network integration. These results imply that children with dyslexia may have quantifiable neuroplasticity as a result of cognitive instruction. EEG-based graph metrics are also highlighted in the paper as possible indicators for tracking dyslexic children’s cognitive development and the effectiveness of their treatment.
- Open Access
- 4 November 2025
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (4) , 045031
Vision-language model approach for few-shot learning of attention deficit hyperactivity disorder using EEG connectivity-based featured images
Traditional medical diagnosis approaches have predominantly relied on single-modality analysis, limiting clinicians to interpreting isolated data streams such as images or time series. The integration of vision language models (VLMs) into neurophysiological analysis represents a paradigm shift toward multimodal diagnostic frameworks, enabling clinicians to interact with diagnosis models through diverse modalities including text, audio, visual inputs, etc. This multimodal interaction capability extends beyond conventional label-based classification, offering clinicians flexibility in diagnostic reasoning and decision-making processes. Building on this foundation, this study explores the application of VLMs to electroencephalography (EEG)-based attention deficit hyperactivity disorder (ADHD) classification, addressing a gap in neurophysiological diagnostics. The proposed framework applies VLM-based few-shot ADHD classification by converting raw EEG data into EEG connectivity-based featured images compatible with contrastive language-image pre-training’s (CLIP) image encoder. The adaptor-based CLIP approach (Tip-Adapter and Tip-Adapter-F) for few-shot learning improves CLIP’s zero-shot classification performance, achieving 78.73% accuracy with 1-shot and 98.30% accuracy with 128-shot using the RN50×16 backbone. Experiments investigate prompt engineering effects, backbone architectures of CLIP, patient-based classification, and combinations of EEG connectivity features. Comparative analysis is performed with two datasets to evaluate the approach between different data sources. Through the adaptation of pre-trained VLMs to neurophysiological data, this technique demonstrates the potential for multimodal diagnostic frameworks that enable flexible clinician-model interactions beyond conventional label-based classification systems. The approach achieves effective ADHD classification with minimal training data while establishing foundations for applying VLMs in clinical neuroscience, where diverse modality interactions through text, visual, and audio inputs can enhance diagnostic workflows. The code is publicly available on GitHub to facilitate further research in the field: https://github.com/miralab-ai/vlm-few-shot-eeg .
- 4 November 2025
- Published by IOP Publishing in Machine Learning: Science and Technology
- Vol. 6 (4) , 045031
Open AccessVision-language model approach for few-shot learning of attention deficit hyperactivity disorder using EEG connectivity-based featured images
Traditional medical diagnosis approaches have predominantly relied on single-modality analysis, limiting clinicians to interpreting isolated data streams such as images or time series. The integration of vision language models (VLMs) into neurophysiological analysis represents a paradigm shift toward multimodal diagnostic frameworks, enabling clinicians to interact with diagnosis models through diverse modalities including text, audio, visual inputs, etc. This multimodal interaction capability extends beyond conventional label-based classification, offering clinicians flexibility in diagnostic reasoning and decision-making processes. Building on this foundation, this study explores the application of VLMs to electroencephalography (EEG)-based attention deficit hyperactivity disorder (ADHD) classification, addressing a gap in neurophysiological diagnostics. The proposed framework applies VLM-based few-shot ADHD classification by converting raw EEG data into EEG connectivity-based featured images compatible with contrastive language-image pre-training’s (CLIP) image encoder. The adaptor-based CLIP approach (Tip-Adapter and Tip-Adapter-F) for few-shot learning improves CLIP’s zero-shot classification performance, achieving 78.73% accuracy with 1-shot and 98.30% accuracy with 128-shot using the RN50×16 backbone. Experiments investigate prompt engineering effects, backbone architectures of CLIP, patient-based classification, and combinations of EEG connectivity features. Comparative analysis is performed with two datasets to evaluate the approach between different data sources. Through the adaptation of pre-trained VLMs to neurophysiological data, this technique demonstrates the potential for multimodal diagnostic frameworks that enable flexible clinician-model interactions beyond conventional label-based classification systems. The approach achieves effective ADHD classification with minimal training data while establishing foundations for applying VLMs in clinical neuroscience, where diverse modality interactions through text, visual, and audio inputs can enhance diagnostic workflows. The code is publicly available on GitHub to facilitate further research in the field: https://github.com/miralab-ai/vlm-few-shot-eeg .
- Open Access
- 25 September 2025
- Published by Academician Publishing
- p. 23-52
Tıbbi Görüntü İşlemede Derin Öğrenme: Teori, Mimariler ve Uygulamalar
Bu bölüm, tıbbi görüntüleme (MR, BT, X-ışını) verilerinin teşhis ve tedavi planlamasındaki kritik rolünü ve bu büyük veri hacmini yorumlamada Derin Öğrenme (DL) yöntemlerinin gerekliliğini incelemektedir. DL'nin, geleneksel Makine Öğreniminden farklı olarak, öznitelik çıkarma ve sınıflandırmayı tek bir süreçte birleştirme yeteneği vurgulanmaktadır. Bölüm, DL'nin temel yapı taşları olan Evrişim, Havuzlama ve Aktivasyon katmanları ile başlıca mimarilerini (özellikle tıbbi segmentasyonda kullanılan U-Net gibi) detaylandırır. Ayrıca, tıbbi görüntülemede veri yetersizliği sorununu çözmek için Transfer Öğrenme stratejisinin önemi ve büyük ölçekli önceden eğitilmiş modellerin (AlexNet, ResNet) bu alana nasıl adapte edileceği de açıklanmıştır. Genel olarak bölüm, tıbbi görüntüleme analizinde DL'nin teorik temelini, mimarilerini ve uygulama yaklaşımlarını kapsamlı bir şekilde sunmaktadır.
- 25 September 2025
- Published by Academician Publishing
- p. 23-52
Open AccessTıbbi Görüntü İşlemede Derin Öğrenme: Teori, Mimariler ve Uygulamalar
Bu bölüm, tıbbi görüntüleme (MR, BT, X-ışını) verilerinin teşhis ve tedavi planlamasındaki kritik rolünü ve bu büyük veri hacmini yorumlamada Derin Öğrenme (DL) yöntemlerinin gerekliliğini incelemektedir. DL'nin, geleneksel Makine Öğreniminden farklı olarak, öznitelik çıkarma ve sınıflandırmayı tek bir süreçte birleştirme yeteneği vurgulanmaktadır. Bölüm, DL'nin temel yapı taşları olan Evrişim, Havuzlama ve Aktivasyon katmanları ile başlıca mimarilerini (özellikle tıbbi segmentasyonda kullanılan U-Net gibi) detaylandırır. Ayrıca, tıbbi görüntülemede veri yetersizliği sorununu çözmek için Transfer Öğrenme stratejisinin önemi ve büyük ölçekli önceden eğitilmiş modellerin (AlexNet, ResNet) bu alana nasıl adapte edileceği de açıklanmıştır. Genel olarak bölüm, tıbbi görüntüleme analizinde DL'nin teorik temelini, mimarilerini ve uygulama yaklaşımlarını kapsamlı bir şekilde sunmaktadır.
- 9 September 2025
- Published by Izmir UOD
- p. 145-152
Recognition of Foot Gestures Through Time-Frequency Analysis and Deep Learning Techniques Using sEMG Data
Most existing approaches for recognizing foot gestures rely on manually extracted features from sEMG recordings. This study aims to develop a deep learning-based method that automatically utilizes the time-frequency representation of sEMG signals from a single-channel sEMG, eliminating the need for manual feature extraction. To achieve this, the continuous wavelet transform (CWT) is used to generate time-frequency RGB images from sEMG signals. Transfer learning is then applied using convolutional neural networks, leveraging pre-recorded foot gesture signal data from healthy individuals. These CWT images are used to classify different foot gestures. The proposed method is evaluated on a dataset of foot gesture sEMG signals, and the results demonstrate that even with single- channel sEMG data, this approach can achieve near state-of-the-art accuracy.
- 9 September 2025
- Published by Izmir UOD
- p. 145-152
Recognition of Foot Gestures Through Time-Frequency Analysis and Deep Learning Techniques Using sEMG Data
Most existing approaches for recognizing foot gestures rely on manually extracted features from sEMG recordings. This study aims to develop a deep learning-based method that automatically utilizes the time-frequency representation of sEMG signals from a single-channel sEMG, eliminating the need for manual feature extraction. To achieve this, the continuous wavelet transform (CWT) is used to generate time-frequency RGB images from sEMG signals. Transfer learning is then applied using convolutional neural networks, leveraging pre-recorded foot gesture signal data from healthy individuals. These CWT images are used to classify different foot gestures. The proposed method is evaluated on a dataset of foot gesture sEMG signals, and the results demonstrate that even with single- channel sEMG data, this approach can achieve near state-of-the-art accuracy.
- 13 August 2024
- Published by SAGE Publications in Transactions of the Institute of Measurement and Control
- Vol. 47 (7) , 1353-1365
Classification of Alzheimer’s dementia EEG signals using deep learning
Alzheimer’s dementia (AD) is a predominant neurological disorder arising from corruptions in brain functions and is characterized by a chronic or progressive nature. While the precise etiology of dementia remains incompletely elucidated, its manifestation is frequently associated with discernible structural and chemical alterations in the brain. Living with dementia significantly impacts individuals’ daily lives due to the resultant loss of cognitive functions. This study presents a novel method to monitor and detect AD using advanced signal processing applied to electroencephalography (EEG) signals. The intrinsic time-scale decomposition (ITD) algorithm is employed to extract proper rotation components (PRCs) from EEG signals, utilizing a 5-second EEG segment duration. The proposed method is compared with the detection of 5-second raw EEG segments using a custom one-dimensional convolutional neural network (1D CNN). Additionally, four different quartiles (Quartile 1 (Q1), Q2, Q3, and Q4) of EEG signals are considered to identify the most significant contributor to AD. Experimental results demonstrate that the ITD-based approach yields better detection performance compared to using raw EEG signals. The most promising result is achieved by the EEG-PRCs method in Q1, with an accuracy of 94.00%, sensitivity of 93.50%, and specificity of 93.90%. In contrast, the highest-performing result of the raw EEG segments method is in Q2, with an accuracy of 88.40%, sensitivity of 89.10%, and specificity of 87.60% in terms of detecting AD.
- 13 August 2024
- Published by SAGE Publications in Transactions of the Institute of Measurement and Control
- Vol. 47 (7) , 1353-1365
Classification of Alzheimer’s dementia EEG signals using deep learning
Alzheimer’s dementia (AD) is a predominant neurological disorder arising from corruptions in brain functions and is characterized by a chronic or progressive nature. While the precise etiology of dementia remains incompletely elucidated, its manifestation is frequently associated with discernible structural and chemical alterations in the brain. Living with dementia significantly impacts individuals’ daily lives due to the resultant loss of cognitive functions. This study presents a novel method to monitor and detect AD using advanced signal processing applied to electroencephalography (EEG) signals. The intrinsic time-scale decomposition (ITD) algorithm is employed to extract proper rotation components (PRCs) from EEG signals, utilizing a 5-second EEG segment duration. The proposed method is compared with the detection of 5-second raw EEG segments using a custom one-dimensional convolutional neural network (1D CNN). Additionally, four different quartiles (Quartile 1 (Q1), Q2, Q3, and Q4) of EEG signals are considered to identify the most significant contributor to AD. Experimental results demonstrate that the ITD-based approach yields better detection performance compared to using raw EEG signals. The most promising result is achieved by the EEG-PRCs method in Q1, with an accuracy of 94.00%, sensitivity of 93.50%, and specificity of 93.90%. In contrast, the highest-performing result of the raw EEG segments method is in Q2, with an accuracy of 88.40%, sensitivity of 89.10%, and specificity of 87.60% in terms of detecting AD.