Browsing by Author "Baygin, Mehmet"

Now showing 1 - 20 of 30

Advancing Pulmonary Embolism Detection with Integrated Deep Learning Architectures
(Springer, 2025) Biret, Can Berk; Gurbuz, Sukru; Akbal, Erhan; Baygin, Mehmet; Ekingen, Evren; Derya, Serdar; Tuncer, Turker
The main aim of this study is to introduce a new hybrid deep learning model for biomedical image classification. We propose a novel convolutional neural network (CNN), named HybridNeXt, for detecting pulmonary embolism (PE) from computed tomography (CT) images. To evaluate the HybridNeXt model, we created a new dataset consisting of two classes: (1) PE and (2) control. The HybridNeXt architecture combines different advanced CNN blocks, including MobileNet, ResNet, ConvNeXt, and Swin Transformer. We specifically designed this model to combine the strengths of these well-known CNNs. The architecture also includes stem, downsampling, and output stages. By adjusting the parameters, we developed a lightweight version of HybridNeXt, suitable for clinical use. To further improve the classification performance and demonstrate transfer learning capability, we proposed a deep feature engineering (DFE) method using a multilevel discrete wavelet transform (MDWT). This DFE model has three main phases: (i) feature extraction from raw images and wavelet bands, (ii) feature selection using iterative neighborhood component analysis (INCA), and (iii) classification using a k-nearest neighbors (kNN) classifier. We first trained HybridNeXt on the training images, creating a pretrained HybridNeXt model. Then, using this pretrained model, we extracted features and applied the proposed DFE method for classification. The HybridNeXt model achieved a test accuracy of 90.14%, while our DFE model improved accuracy to 96.35%. Overall, the results confirm that our HybridNeXt architecture is highly accurate and effective for biomedical image classification. The presented HybridNeXt and HybridNeXt-based DFE methods can potentially be applied to other image classification tasks.
Automated Anxiety Detection Using Probabilistic Binary Pattern with ECG Signals
(Elsevier Ireland Ltd, 2024) Baygin, Mehmet; Barua, Prabal Datta; Dogan, Sengul; Tuncer, Turker; Hong, Tan Jen; March, Sonja; Acharya, U. Rajendra
Background and aim: Anxiety disorder is common; early diagnosis is crucial for management. Anxiety can induce physiological changes in the brain and heart. We aimed to develop an efficient and accurate handcrafted feature engineering model for automated anxiety detection using ECG signals. Materials and methods: We studied open-access electrocardiography (ECG) data of 19 subjects collected via wearable sensors while they were shown videos that might induce anxiety. Using the Hamilton Anxiety Rating Scale, subjects are categorized into normal, light anxiety, moderate anxiety, and severe anxiety groups. ECGs were divided into non-overlapping 4- (Case 1), 5- (Case 2), and 6-second (Case 3) segments for analysis. We proposed a self-organized dynamic pattern-based feature extraction function-probabilistic binary pattern (PBP)- in which patterns within the function were determined by the probabilities of the input signal-dependent values. This was combined with tunable q-factor wavelet transform to facilitate multileveled generation of feature vectors in both spatial and frequency domains. Neighborhood component analysis and Chi2 functions were used to select features and reduce data dimensionality. Shallow k-nearest neighbors and support vector machine classifiers were used to calculate four (=2 x 2) classifier-wise results per input signal. From the latter, novel selforganized combinational majority voting was applied to calculate an additional five voted results. The optimal final model outcome was chosen from among the nine (classifier-wise and voted) results using a greedy algorithm. Results: Our model achieved classification accuracies of over 98.5 % for all three cases. Ablation studies confirmed the incremental accuracy of PBP-based feature engineering over traditional local binary pattern feature extraction. Conclusions: The results demonstrated the feasibility and accuracy of our PBP-based feature engineering model for anxiety classification using ECG signals.
Automated Detection of Gastrointestinal Diseases Using Resnet50*-Based Explainable Deep Feature Engineering Model with Endoscopy Images
(MDPI, 2024) Cambay, Veysel Yusuf; Barua, Prabal Datta; Hafeez Baig, Abdul; Dogan, Sengul; Baygin, Mehmet; Tuncer, Turker; Acharya, U. R.
This work aims to develop a novel convolutional neural network (CNN) named ResNet50* to detect various gastrointestinal diseases using a new ResNet50*-based deep feature engineering model with endoscopy images. The novelty of this work is the development of ResNet50*, a new variant of the ResNet model, featuring convolution-based residual blocks and a pooling-based attention mechanism similar to PoolFormer. Using ResNet50*, a gastrointestinal image dataset was trained, and an explainable deep feature engineering (DFE) model was developed. This DFE model comprises four primary stages: (i) feature extraction, (ii) iterative feature selection, (iii) classification using shallow classifiers, and (iv) information fusion. The DFE model is self-organizing, producing 14 different outcomes (8 classifier-specific and 6 voted) and selecting the most effective result as the final decision. During feature extraction, heatmaps are identified using gradient-weighted class activation mapping (Grad-CAM) with features derived from these regions via the final global average pooling layer of the pretrained ResNet50*. Four iterative feature selectors are employed in the feature selection stage to obtain distinct feature vectors. The classifiers k-nearest neighbors (kNN) and support vector machine (SVM) are used to produce specific outcomes. Iterative majority voting is employed in the final stage to obtain voted outcomes using the top result determined by the greedy algorithm based on classification accuracy. The presented ResNet50* was trained on an augmented version of the Kvasir dataset, and its performance was tested using Kvasir, Kvasir version 2, and wireless capsule endoscopy (WCE) curated colon disease image datasets. Our proposed ResNet50* model demonstrated a classification accuracy of more than 92% for all three datasets and a remarkable 99.13% accuracy for the WCE dataset. These findings affirm the superior classification ability of the ResNet50* model and confirm the generalizability of the developed architecture, showing consistent performance across all three distinct datasets.
An Automated Earthquake Classification Model Based on a New Butterfly Pattern Using Seismic Signals
(Pergamon-Elsevier Science Ltd, 2024) Ozkaya, Suat Gokhan; Baygin, Mehmet; Barua, Prabal Datta; Tuncer, Turker; Dogan, Sengul; Chakraborty, Subrata; Acharya, U. Rajendra
Background: Seismic signals are useful for earthquake detection and classification. Therefore, various artificial intelligence (AI) models have been used with seismic signals to develop automated earthquake detection systems. Our primary goal is to present an accurate feature engineering model for earthquake detection and classification using seismic signals.Material and model: We have used a public dataset in this work containing three categories: (1) noise, (2) P waves, and (3) S waves. P and S waves are used to define earthquakes. We have presented two applied use cases using this dataset: (i) earthquake detection and (ii) wave classification. In this work, a new textural feature extractor has been presented by using a graph pattern similar to a butterfly. Thus, this feature extraction function is named Butterfly pattern (BFPat). We have created a new feature engineering architecture by deploying BFPat, statistics, and wavelet packet decomposition (WPD) functions. The recommended BFPat and statistics have been applied to the wavelet bands created by WPD and the raw seismic signals. Multilevel features have been extracted from both frequency and space domains. The used dataset contains signals with three channels. Using these three channels, seven signals have been created. Seven feature vectors have been created from 7 input signals used in this study. The most meaningful/informative features from the generated feature set are then selected using the iterative neighborhood component analysis feature selector method. Seven chosen feature vectors have been considered as inputs of the two shallow classifiers: k nearest neighbors (kNN) and support vector machine (SVM). A total of 14 (=7 x 2) results have been obtained in the classification phase. A majority voting process was applied in the last phase to choose the best results and improve the classification performance.Results: We have presented two use cases for our new BFPat method in this work to obtain superior results. Our model reached an accuracy of 99.58% in detecting the earthquake detection and 93.13% accuracy in 3-class classifications of waves.Conclusions: Our recommended model has achieved over 90% classification performance for both cases. Also, we have presented the most valuable channel and combinations in our work. Our developed system is ready to be tested with a bigger database.
Automated Mental Arithmetic Performance Detection Using Quantum Pattern- and Triangle Pooling Techniques with EEG Signals
(Pergamon-Elsevier Science Ltd, 2023) Baygin, Nursena; Aydemir, Emrah; Barua, Prabal D.; Baygin, Mehmet; Doganm, Sengul; Tuncer, Turker; Acharya, U. Rajendra
Background: Electroencephalography (EEG) signals recorded during mental arithmetic tasks can be used to quantify mental performance. The classification of these input EEG signals can be automated using machine learning models. We aimed to develop an efficient handcrafted model that could accurately discriminate "bad counters" vs. "good counters" in mental arithmetic. Materials and method: We studied a public mental arithmetic task performance EEG dataset comprising 20-channel EEG signal segments recorded from 36 healthy right-handed subjects divided into two classes 10 "bad counters" and 26 "good counters". The original 60-second EEG samples are divided into 424 15-second segments (119 and 305 belonging to the "bad counters" and "good counters", respectively) to input into our model. Our model comprised a novel multilevel feature extraction method based on (1) four rhombuses lattice pattern, a new generation function for feature extraction that was inspired by the lattice structure in post-quantum cryptography; and (2) triangle pooling, a new distance-based pooling function for signal decomposition. These were combined with downstream feature selection using iterative neighborhood component analysis, channel-wise result classification using support vector machine with leave-one-subject-out (LOSO) and 10-fold) crossvalidations (CVs) to calculate prediction vectors, iterative majority voting to generate voted vectors, and greedy algorithm to obtain the best results. Results: The model attained 88.44% and 96.42% geometric means and accuracies of 93.40% and 97.88%, using LOSO and 10-fold CVs, respectively. Conclusions: Our model's >93% classification accuracies compared favorably against published literature. Importantly, the model has linear computational complexity, which enhances its ease of implementation.
Black-White Hole Pattern: An Investigation on the Automated Chronic Neuropathic Pain Detection Using Eeg Signals
(Springer, 2024) Tasci, Irem; Baygin, Mehmet; Barua, Prabal Datta; Hafeez-Baig, Abdul; Dogan, Sengul; Tuncer, Turker; Acharya, U. Rajendra
Electroencephalography (EEG) signals provide information about the brain activities, this study bridges neuroscience and machine learning by introducing an astronomy-inspired feature extraction model. In this work, we developed a novel feature extraction function, black-white hole pattern (BWHPat) which dynamically selects the most suitable pattern from 14 options. We developed BWHPat in a four-phase feature engineering model, involving multileveled feature extraction, feature selection, classification, and cortex map generation. Textural and statistical features are extracted in the first phase, while tunable q-factor wavelet transform (TQWT) aids in multileveled feature extraction. The second phase employs iterative neighborhood component analysis (INCA) for feature selection, and the k-nearest neighbors (kNN) classifier is applied for classification, yielding channel-specific results. A new cortex map generation model highlights the most active channels using median and intersection functions. Our BWHPat-driven model consistently achieved over 99% classification accuracy across three scenarios using the publicly available EEG pain dataset. Furthermore, a semantic cortex map precisely identifies pain-affected brain regions. This study signifies the contribution to EEG signal classification and neuroscience. The BWHPat pattern establishes a unique link between astronomy and feature extraction, enhancing the understanding of brain activities.
Deep Learning in Forensic Analysis: Optical Coherence Tomography Image Classification in Methamphetamine Detection
(Pergamon-Elsevier Science Ltd, 2025) Gurbuzer, Nilifer; Ozkaya, Alev Lazoglu; Yaylali, Elif Topdagi; Tozoglu, Elif Ozcan; Baygin, Mehmet; Tasci, Burak; Tuncer, Turker
Detecting drug addiction in forensic science traditionally relies on expensive and time-consuming laboratory tests. This study proposes a rapid, non-invasive approach that uses optical coherence tomography images combined with deep learning techniques to identify methamphetamine users. A novel convolutional neural network was developed, incorporating depthwise and pointwise convolutions, patchify-based downsampling, and inception blocks to improve feature extraction and classification accuracy. To further enhance model performance, we introduced a grid-based deep feature engineering model that extracts and selects discriminative features using iterative neighborhood component analysis. The proposed model achieved 91.02 % accuracy, surpassing the 88.57 % accuracy of Mobile Network version 2 on the same dataset. By integrating the grid-based feature engineering model, classification accuracy was further improved to 93.27 %, demonstrating a significant enhancement over traditional deep learning approaches. The dataset consisted of 2172 optical coherence tomography images collected from 54 methamphetamine users and 60 control subjects, ensuring a diverse and representative sample. This research marks the first application of optical coherence tomography imaging in drug addiction detection, bridging biomedical imaging and forensic science. By employing gradient-weighted class activation mapping visualization, we identified key retinal features that distinguish methamphetamine users from non-users, thereby making the model more interpretable and clinically relevant. Given its high accuracy, lightweight architecture, and non-invasive nature, the proposed method offers a promising forensic tool for rapid, artificial intelligence-driven drug addiction screening with potential real-world applicability in forensic investigations and healthcare.
Directed Lobish-Based Explainable Feature Engineering Model with Ttpat and Cwinca for Eeg Artifact Classification
(Elsevier, 2024) Tuncer, Turker; Dogan, Sengul; Baygin, Mehmet; Tasci, Irem; Mungen, Bulent; Tasci, Burak; Acharya, U. R.
Background and Objective: Electroencephalography (EEG) signals are crucial to decipher various brain activities. However, these EEG signals are subtle and contain various artifacts, which can happen due to various reasons. The main aim of this paper is to develop an explainable novel machine learning model that can identify the cause of these artifacts. Material and method: A new EEG signal dataset was collected to classify various types of artifacts. This dataset contains eight classes: seven are artifacts, and one is the EEG signal without artifacts. A novel feature engineering model has been proposed to classify these artifact classes automatically. This model contains three main steps: (i) feature generation with the proposed transition table pattern (TTPat), (ii) the proposed cumulative weight-based iterative neighborhood component analysis (CWINCA)-based feature selection, and (iii) classification using t algorithm-based k-nearest neighbors (tkNN). The novelty of this work is TTPat feature extractor and CWINCA feature selector. Channel-based transformation is performed using the proposed TTPat, which extracts 392 features from the transformed EEG signal. A novel CWINCA feature selector is proposed. The artifacts are classified using tkNN algorithm. Results: The proposed TTPat and CWINCA-based feature engineering model obtained a classification accuracy ranging from 66.39% to 97.69% for 30 cases. We presented the explainable results using a new symbolic language termed Directed Lobish. Conclusions: The results and findings demonstrated that the proposed explainable feature engineering (EFE) model is good at artifact detection and classification. Directed Lobish has been presented to obtain explainable results and is a new symbolic language.
DSWIN: Automated Hunger Detection Model Based on Hand-Crafted Decomposed Shifted Windows Architecture Using EEG Signals
(Elsevier, 2024) Kirik, Serkan; Tasci, Irem; Barua, Prabal D.; Yildiz, Arif Metehan; Keles, Tugce; Baygin, Mehmet; Acharya, U. R.
Hunger is a physiological state that arises from complex interactions of multiple factors, including higher brain center control. We purposed to develop an accurate and efficient machine-learning model for the automated detection of hunger using EEG signals. We prospectively acquired 14-channel EEG (sampling frequency 128 Hz) from 43 and 48 fasted and post-prandial healthy subjects (hungry vs. control groups, respectively) using the EMOTIV EPOC+ mobile brain cap system. To augment the hunger response, fasted subjects were also shown video images of food during EEG recording. The EEG signals were divided into 15-second segments. 877 and 852 participants/subjects were in the hungry and control groups. We created a novel handcrafted architecture-decomposed shifted window (DSWIN)-that combined swin patch division with tunable Q-factor wavelet transform-based signal decomposition for multilevel feature extraction of EEG signals. Textural and statistical features were extracted from the multiple patches and decomposed signals using a novel penta pattern-based extractor and statistical moments, respectively, and then merged. Iterative neighborhood component analysis (INCA) and iterative ReliefF (IRF) were applied. Twenty-eight selected feature vectors were generated, which were then fed to a shallow k-nearest neighbors (kNN) classifier to calculate channel-wise prediction vectors. From the 28 channel-wise prediction vectors, another 26 modes of function-based voted results were calculated using iterative hard majority voting, and the best overall model result was selected using a greedy algorithm. Our model attained 99.54% and 82.71% binary classification accuracies of hungry status vs. control using 10-fold and leave-one-subject-out cross-validations, respectively.
Efdensenet: Automated Pulmonary Hypertension Detection Model Based on Efficientnetb0 and Densenet201 Using CT Images
(IEEE-Inst Electrical Electronics Engineers Inc, 2023) Kivrak, Tarik; Nayak, Jagadish; Gelen, Mehmet Ali; Barua, Prabal Datta; Baygin, Mehmet; Pamukcu, Hilal Erken; Acharya, U. Rajendra
Pulmonary hypertension (PH) is a chronic and progressive disease. We introduced a novel automated self-organized feature engineering architecture for PH detection, which was trained and refined using a new thoracic CT image dataset. This study's dataset includes 807 transverse contrast-enhanced CT images from 313 patients, categorized into four groups: Group 1 with 20 mmHg 30 mmHg; and a control group with no PH. Our model consists of four primary stages: (i) generation of features based on combinations from nested patches, (ii) feature selection, (iii) classification and (iv) majority voting. CT images were segmented into nested patches, each being processed through pretrained EfficientNetB0 and DenseNet201 to derive four deep feature vectors, utilizing both the global average pooling and fully connected layers of these networks. These four extracted features underwent combinatorial operations, resulting in 15 feature vectors. Subsequently, these vectors were introduced to neighborhood component analysis, ReliefF, and Chi2 feature selectors. This process yielded 45 refined feature vectors with diminished data dimensions. These selected vectors were then processed through a support vector machine and k-nearest neighbors classifiers, producing 90 predictive vectors. By applying mode-based iterative majority voting to these vectors, an additional 88 voted prediction vectors were generated, leading to a total of 178 classifier-generated and voted prediction vectors. The optimal classification result was selected from these 178 vectors. With the use of 10-fold cross-validation, our model achieved a remarkable 97.27% overall accuracy for the 4-class classification on the study dataset. Owing to its reduced time complexity, this model is practical for CT-based PH screenings.
Exdarklbp: A Hybrid Deep Feature Generation-Based Genetic Malformation Detection Using Facial Images
(Springer, 2023) Barua, Prabal Datta; Kirik, Serkan; Dogan, Sengul; Koc, Canan; Ozkaynak, Fatih; Baygin, Mehmet; Acharya, U. Rajendra
Body malformations, including those affecting the face, can arise as a result of genetic disorders. The diagnosis of such changes may often require specialist expertise, which is scarce. In this study, we have presented a computer vision model capable of accurately classifying malformed vs. non-malformed face images using automated classification techniques. Our model, which we refer to as ExDarkLBP (exemplar/patch-based feature extraction deploying pretrained DarkNet and local binary pattern), is based on exemplar hybrid feature engineering and incorporates two primary feature extraction methods: (i) textural feature generation using local binary pattern (LBP) and (ii) deep feature creation deploying pretrained DarkNet53. The most informative 500 textural and 500 deep features were first selected using the neighborhood component analysis (NCA) feature selection function and then merged to form a 1000 feature vector. This vector was subsequently fed to iterative NCA to choose the most valuable features. By combining this optimal feature vector with a support vector machine, we achieved an accuracy of 99.22% using a ten-fold cross-validation strategy. Our proposed ExDarkLBP model is highly accurate and may be potentially applied for the screening of facial malformations associated with genetic disorders using face images.
Explainable Attention ResNet18-Based Model for Asthma Detection Using Stethoscope Lung Sounds
(Pergamon-Elsevier Science Ltd, 2023) Topaloglu, Ihsan; Barua, Prabal Datta; Yildiz, Arif Metehan; Keles, Tugce; Dogan, Sengul; Baygin, Mehmet; Acharya, U. Rajendra
This study proposes an accurate asthma detection model using an attention network and machine learning technique. The objective of this study is the automated detection of asthma using an attention network. The lung sounds from 203 subjects involving 767 segments from asthma and 722 segments from healthy subjects were collected using a stethoscope. A novel Attention ResNet18-based deep feature engineering model has been developed in five phases: preprocessing, training the Attention ResNet18 network, extracting deep features, iterative feature selection, and classification using k-nearest neighbor (kNN) or support vector machine (SVM). Gradient-weighted class activation mapping (Grad-CAM) was used to generate heat maps, effectively distinguishing asthma lung sounds from those of normal individuals. By using Grad-CAM, explainable results have been presented. Our proposed model obtained an accuracy of 99.73% using SVM with 10-fold cross-validation, surpassing the performance obtained by previous models. Hence, the developed model has the potential to detect asthma in the real-world scenario. scenario to detect.
FGPat18: Feynman Graph Pattern-Based Language Detection Model Using EEG Signals
(Elsevier Sci Ltd, 2023) Kirik, Serkan; Dogan, Sengul; Baygin, Mehmet; Barua, Prabal Datta; Demir, Caner Feyzi; Keles, Tugce; Acharya, U. Rajendra
We aimed to develop an efficient handcrafted feature engineering model based on four directed graphs modeled on Feynman graph patterns (FGPat) for electroencephalography (EEG)-based language identification. We prospectively acquired a 3252-EEG dataset from 20 native English-speaking Nigerian-born and 20 Turkish subjects who were shown 20 standardized sentences in the English and Turkish languages, respectively. 14-channel 15-second EEG signals (sampling frequency 128 Hz) were acquired using the EMOTIV EPOC+ mobile brain cap system. In our FGPat18 model, input EEG signals and their 17 tunable Q wavelet transform-decomposed wavelet bands were fed as input to four FGPat-based feature extraction functions and statistical feature generators to extract textural and statistical features, respectively. Then they were concatenated to obtain four final feature vectors of varying lengths. The latter was input to the neighborhood component analysis function to select the most discriminative/meaningful 256 vectors in each vector, which were then fed to the k-nearest neighbor (kNN) classifier for binary classification. Next, iterative majority voting (IMV) was applied to the four kNN-predicted vectors to generate two voted vectors; the most accurate among the six pooled vectors was then selected as the best channel-wise result. Finally, all 14 channel-wise best vectors were input to the IMV algorithm again to calculate another 12 voted vectors; the best overall result for the EEG study was chosen among the 26 vectors. FGPat18 attained 99.38% and 92.47% classification accuracy rates with 10-fold and leave-one-subject-out cross-validations, respectively. The model has linear complexity.
Flexilpq: Automated Osteoid Osteoma Detection Using Computed Tomography
(Springer London Ltd, 2026) Key, Sefa; Agar, Anil; Sercek, Ilknur; Poyraz, Ahmet Kursad; Baygin, Mehmet; Dogan, Sengul; Tuncer, Turker
In this study, we introduce FlexiLPQ, an innovative image classification model developed as the feature-engineering counterpart of FlexiViT, and evaluate its performance on Osteoid Osteoma diagnosis. For this purpose, a new Osteoid Osteoma CT dataset was curated. Using this dataset, our goal was to design an automatic detection system capable of identifying Osteoid Osteoma with high accuracy. The proposed FlexiLPQ model operates through five main phases: (i) multi-patch feature extraction using local phase quantization (LPQ), (ii) feature selection via cumulative weighted iterative neighborhood component analysis (CWINCA), (iii) classification using a t-algorithm-based k-nearest neighbors (tkNN) classifier, (iv) iterative majority voting (IMV) to refine the decision, and (v) selection of the best final outcome. FlexiLPQ was applied to the curated CT dataset and achieved a 98.89% classification accuracy. Additionally, multiple patch sizes were incorporated, and their performance differences were analyzed. The results clearly show that FlexiLPQ is an effective and robust image classification framework, and it is well suited for biomedical imaging tasks such as Osteoid Osteoma detection.
Flower Automata Pattern-Based Discrimination of Fibromyalgia from Control Subjects Using Fusion of Sleep EEG and ECG Signals
(IEEE-Inst Electrical Electronics Engineers Inc, 2025) Barua, Prabal Datta; Kobayashi, Makiko; Dogan, Sengul; Baygin, Mehmet; Tuncer, Turker; Paul, Jose Kunnel; Acharya, U. R.
Electroencephalogram (EEG) and electrocardiogram (ECG) signals provide vital insights into brain and heart activity and are widely used in automated medical diagnostics. This study introduces a novel, multimodal fibromyalgia detection system developed by the fusion of EEG and ECG signals recorded during sleep stages 2 and 3. The novelty of the model is the use of dynamic and interpretable feature engineering framework comprising of two innovations: 1) Flower Automata Pattern (FAP) for self-organized pattern-based feature extraction, and 2) Attention-Driven Wavelet Transform and Absolute Maximum Pooling (ADWTAMP) method for signal decomposition and compression. Three feature selection strategies-Neighborhood Component Analysis (NCA), Chi2, and the intersection of NCA and Chi2 (NCAChi2) - are employed to generate robust feature vectors, which are classified using k-nearest neighbors (kNN) and support vector machine (SVM) under the leave-one-record-out cross-validation (LORO CV) scheme. The final decision is derived through an iterative voting and greedy fusion approach. The proposed model achieved classification accuracies of 99.36% and 98.37% for sleep stages 2 and 3, respectively. Key advantages of the model include its high accuracy, low computational requirements (CPU-only execution), and explainable architecture. To the best of our knowledge, this is the first multimodal automata-based classification framework designed for fibromyalgia detection.
INCR: Inception and Concatenation Residual Block-Based Deep Learning Network for Damaged Building Detection Using Remote Sensing Images
(Elsevier, 2023) Tasci, Burak; Acharya, Madhav R.; Baygin, Mehmet; Dogan, Sengul; Tuncer, Turker; Belhaouari, Samir Brahim
In February 2023, Turkey experienced a series of earthquakes that caused significant damage to buildings and affected many people. Detecting building damage quickly is crucial for helping earthquake victims, and we believe machine learning models offer a promising solution. In our research, we introduce a new, lightweight deep-learning model capable of accurately classifying damaged buildings in remote-sensing datasets. Our main goal is to create an automated damage detection system using a novel deep-learning model. We started by collecting a new dataset with two categories: damaged and undamaged buildings. Then, we developed a unique convolutional neural network (CNN) called the inception and concatenation residual (InCR) deep learning network, which incorporates concatenation-based residual blocks and inception blocks to improve performance. We trained our InCR model on the newly collected dataset and used it to extract features from images using global average pooling. To refine these features and select the most informative ones, we applied iterative neighborhood component analysis (INCA). Finally, we classified the refined features using commonly used shallow classifiers. To evaluate our method, we used tenfold cross-validation (10-fold CV) with eight classifiers. The results showed that all classifiers achieved classification accuracies higher than 98 %. This demonstrates that our proposed InCR model is a viable option for CNNs and can be used to create an accurate automated damage detection application. Our research presents a unique solution to the challenge of automated damage detection after earthquakes, showing promising results that highlight the potential of our approach.
Innovative Fibromyalgia Detection Approach Based on Quantum-Inspired 3lbp Feature Extractor Using ECG Signal
(IEEE-Inst Electrical Electronics Engineers Inc, 2023) Barua, Prabal Datta; Kobayashi, Makiko; Tanabe, Masayuki; Baygin, Mehmet; Paul, Jose Kunnel; Iype, Thomas; Acharya, U. Rajendra
Fibromyalgia is a chronic pain syndrome associated with sleep disturbances, which may manifest as altered electroencephalography and electrocardiography (ECG) signal alterations during sleep. We aimed to develop a lightweight machine learning model for diagnosing fibromyalgia using single-lead ECG signals recorded during sleep. We analyzed 139 single-lead ECGs recorded during Stage 2 and Sleep Stage 3 of 16 patients with fibromyalgia and 16 age and sex matched controls. ECG records were divided into 15-second segments: 3308 and 1783 in healthy vs fibromyalgia classes, respectively. Our model comprised (1) feature extraction that combined an 8-wavelet filter and 4-level multiple filters-based multilevel discrete wavelet transform signal decomposition with a novel local binary pattern (LBP)-like function, 3LBP, that generated multiple patterns (analogous to quantum superposition) for feature map value extraction (the optimal input-specific pattern was dynamically selected using a novel forward-forward algorithm); (2) feature selection using neighborhood component analysis and Chi-square functions; (3) classification with k-nearest neighbors and support vector machine classifiers using leave-one-record-out cross-validation; and (4) mode function-based iterative majority voting to generate voted results, from which the best model result was derived. Our model attained binary classification accuracies of 93.87% and 92.02% for Sleep Stage 2 and Sleep Stage 3, respectively. The observed outcomes and empirical evidence unequivocally demonstrate the efficacy of our proposed methodology in differentiating the electrocardiographic signatures of fibromyalgia patients from control subjects. The model exhibited self-organizational properties and computational efficiency, rendering it amenable to facile clinical integration.
Lattice 123 Pattern for Automated Alzheimer's Detection Using EEG Signal
(Springer, 2024) Dogan, Sengul; Barua, Prabal Datta; Baygin, Mehmet; Tuncer, Turker; Tan, Ru-San; Ciaccio, Edward J.; Acharya, U. Rajendra
This paper presents an innovative feature engineering framework based on lattice structures for the automated identification of Alzheimer's disease (AD) using electroencephalogram (EEG) signals. Inspired by the Shannon information entropy theorem, we apply a probabilistic function to create the novel Lattice123 pattern, generating two directed graphs with minimum and maximum distance-based kernels. Using these graphs and three kernel functions (signum, upper ternary, and lower ternary), we generate six feature vectors for each input signal block to extract textural features. Multilevel discrete wavelet transform (MDWT) was used to generate low-level wavelet subbands. Our proposed model mirrors deep learning approaches, facilitating feature extraction in frequency and spatial domains at various levels. We used iterative neighborhood component analysis to select the most discriminative features from the extracted vectors. An iterative hard majority voting and a greedy algorithm were used to generate voted vectors to select the optimal channel-wise and overall results. Our proposed model yielded a classification accuracy of more than 98% and a geometric mean of more than 96%. Our proposed Lattice123 pattern, dynamic graph generation, and MDWT-based multilevel feature extraction can detect AD accurately as the proposed pattern can extract subtle changes from the EEG signal accurately. Our prototype is ready to be validated using a large and diverse database.
Lobish: Symbolic Language for Interpreting Electroencephalogram Signals in Language Detection Using Channel-Based Transformation and Pattern
(MDPI, 2024) Tuncer, Turker; Dogan, Sengul; Tasci, Irem; Baygin, Mehmet; Barua, Prabal Datta; Acharya, U. Rajendra
Electroencephalogram (EEG) signals contain information about the brain's state as they reflect the brain's functioning. However, the manual interpretation of EEG signals is tedious and time-consuming. Therefore, automatic EEG translation models need to be proposed using machine learning methods. In this study, we proposed an innovative method to achieve high classification performance with explainable results. We introduce channel-based transformation, a channel pattern (ChannelPat), the t algorithm, and Lobish (a symbolic language). By using channel-based transformation, EEG signals were encoded using the index of the channels. The proposed ChannelPat feature extractor encoded the transition between two channels and served as a histogram-based feature extractor. An iterative neighborhood component analysis (INCA) feature selector was employed to select the most informative features, and the selected features were fed into a new ensemble k-nearest neighbor (tkNN) classifier. To evaluate the classification capability of the proposed channel-based EEG language detection model, a new EEG language dataset comprising Arabic and Turkish was collected. Additionally, Lobish was introduced to obtain explainable outcomes from the proposed EEG language detection model. The proposed channel-based feature engineering model was applied to the collected EEG language dataset, achieving a classification accuracy of 98.59%. Lobish extracted meaningful information from the cortex of the brain for language detection.
Mnpdensenet: Automated Monkeypox Detection Using Multiple Nested Patch Division and Pretrained Densenet201
(Springer, 2024) Demir, Fahrettin Burak; Baygin, Mehmet; Tuncer, Ilknur; Barua, Prabal Datta; Dogan, Sengul; Tuncer, Turker; Acharya, U. Rajendra
BackgroundMonkeypox is a viral disease caused by the monkeypox virus (MPV). A surge in monkeypox infection has been reported since early May 2022, and the outbreak has been classified as a global health emergency as the situation continues to worsen. Early and accurate detection of the disease is required to control its spread. Machine learning methods offer fast and accurate detection of COVID-19 from chest X-rays, and chest computed tomography (CT) images. Likewise, computer vision techniques can automatically detect monkeypoxes from digital images, videos, and other inputs.ObjectivesIn this paper, we propose an automated monkeypox detection model as the first step toward controlling its global spread.Materials and methodA new dataset comprising 910 open-source images classified into five categories (healthy, monkeypox, chickenpox, smallpox, and zoster zona) was created. A new deep feature engineering architecture was proposed, which contained the following components: (i) multiple nested patch division, (ii) deep feature extraction, (iii) multiple feature selection by deploying neighborhood component analysis (NCA), Chi2, and ReliefF selectors, (iv) classification using SVM with 10-fold cross-validation, (v) voted results generation by deploying iterative hard majority voting (IHMV) and (vi) selection of the best vector by a greedy algorithm.ResultsOur proposal attained a 91.87% classification accuracy on the collected dataset. This is the best result of our presented framework, which was automatically selected from 70 generated results.ConclusionsThe computed classification results and findings demonstrated that monkeypox could be successfully detected using our proposed automated model.