BSc Computer engineering / University of Baghdad
MSc Computer Network and Communication /UPM
RESEARCH INTERESTS
Computer Network , WSN , IOV , SDVN , FoG computing and VANET
13
Scopus Publications
Scopus Publications
Transformer Network on Global Self-Attention Mechanism for Brain Tumor Segmentation Ammar Awni Abbas Baghdadi, Mohammed Al-Mukhtar, Ammar S Al-Zubaidi Baghdad Science Journal, 2025 Transformers are a specific category of neural network design. Transformers often depend on extensive pre-training on a large scale and exhibit a notable degree of computational complexity. The disadvantage of using this method is a significant increase in computational complexity, which necessitates a significant commitment of time and computing resources in order to successfully work with these models. Transformer networks possess the desirable benefit of extracting distant characteristics effectively via their self-attention mechanism. In this paper, the Global Self-Attention Transformer module is applied to tackle these issues. The model is based on a segmentation problem called Brain-GS that works as a mechanism and encompasses several forms, one of which is global self-attention. The aim of the experiment is to attain the best precision in segmentation lesions. Unlike localized self-attention, global self-attention assigns equal importance to all items within a given sequence. Global attention mechanism was used that demonstrates high efficiency Unet, making it suitable as the fundamental component of a deep neural network. The model is able to comprehend and accurately reflect the long-range relationships that are present in the data. Using the densnet and Resnet50 backbones, our approach is compared to the recommended architecture in the context of multimodal brain tumor segmentation. The proposed models may have a big effect on the prognosis and treatment of people with glioblastoma, a type of brain cancer that is very likely to be fatal. Our own model achieved a 0.896 dice score and an accuracy of 0.987, and Jaccard achieved 0.901 for validation data and tumor core.
DAB-UNET: Dual Attention Block UNET Segmentation for Diabetic Retinopathy Utilizing an Encoder-Decoder Residual Haithem Kareem Abass Journal of Image and Graphics United Kingdom, 2025 —Fundus images play an essential role in ophthalmic diagnostics for the detection of many eye illnesses. The experiment begins with a thorough image pre-processing technique, which includes clipping the circular borders, scaling the image, enhancing the contrast, removing noise, and augmenting the data. The new combined block applies to extracting distinctive deep feature representations, which help to detect the first shape of the edges of each lesion. It is namely the Attention Block and the Conv-Deconv UNET model. Attention Block is subsequently implemented in order to augment the robustness and quality of feature depictions derived from a pair of DR images. The Dual Attention Block for the backbone, which is supplemented with hierarchical bottleneck attention, is what we propose here referred to as Dual Attention Block UNET (DAB-UNET). Bottleneck Attention Blocks and Dual Attention Blocks greatly improve a model’s ability to concentrate on essential features, boosting its performance in complex tasks such as image segmentation. When these attention mechanisms are built into architectures like DAB-UNET, they make the network faster and more accurate, letting it pick up on small, specific details. This is particularly beneficial in areas like medical imaging, where high precision is essential. In order to emphasize retinal anomalies that are significant for fovea macula and Diabetic Retinopathy (DR) semantic segmentation in the deteriorated retina, the network is made up of a unique bottleneck attention block. We trained Mask-Region based Convoluting Neural Network (RCNN) model that comprises of a backbone for eliminating Oculus Dexter (OD) regions. Moreover, the proposed block combines self-attention with channel attention in order to highlight these abnormalities. Our results indicate that DAB-UNET is potentially very effective for identifying landmarks even when dealing with different types of retinal degenerative disorders.
Arrhythmia Recognition Algorithm Using Squared Krawtchouk-Tchebichef Polynomial-Based Feature Extraction Ammar S. Al-Zubaidi, Raafat Salih Muhammad, Hayder S. Radeaf, Basheera M Mahmmod, Sadiq H. Abdulhussain, Muntadher Alsabah, Abir Hussain Proceedings 18th International Conference on Developments in Esystems Engineering Dese 2025, 2025 Accurate and efficient recognition of cardiac arrhythmias is crucial for fast diagnosis and accurate prevention of cardiovascular diseases. This paper investigates the use of an electrocardiogram (ECG) signal analysis based on hybrid form of orthogonal polynomials, namely, Squared Krawtchouk-Tchebichef Polynomials (SKTP) to obtain accurate recognition of cardiac arrhythmias. The proposed method exploits the combined properties representation power of Krawtchouk and Tchebichef moments to capture discriminative characteristics of ECG waveforms. The SKTP has the ability to enhance the feature robustness against noise and baseline wander. ECG signals from the compiled MIT-BIH arrhythmia dataset are used for performance evaluation of the proposed method. SKTP-based features are then computed from each segment and fed into Support Vector Machine (SVM) for arrhythmia classification. Experimental results demonstrate that the proposed method achieves high recognition performance compared to the existing works, attaining an overall accuracy of 91.33%. The findings indicate that SKTP-based features provide a compact, and discriminative representation of ECG signals, making them suitable for real-time arrhythmia monitoring in wearable and telemedicine applications.
Skin Lesion Segmentation for Melanoma Using Dilated DenseUNet Ammar Al-Zubaidi, Mohammed Al-Mukhtar, Mina H. Al-hashimi, Haris Ijaz Journal of ICT Research and Applications, 2024 Melanoma, a highly malignant form of skin cancer, affects individuals of all genders and is associated with high mortality rates, especially in advanced stages. The use of tele-dermatology has emerged as a proficient diagnostic approach for skin lesions and is particularly beneficial in rural areas with limited access to dermatologists. However, accurately, and efficiently segmenting melanoma remains a challenging task due to the significant diversity observed in the morphology, pigmentation, and dimensions of cutaneous nevi. To address this challenge, we propose a novel approach called DenseUNet-169 with a dilated convolution encoder-decoder for automatic segmentation of RGB dermascopic images. By incorporating dilated convolution, our model improves the receptive field of the kernels without increasing the number of parameters. Additionally, we used a method called Copy and Concatenation Attention Block (CCAB) for robust feature computation. To evaluate the performance of our proposed framework, we utilized the International Skin Imaging Collaboration (ISIC) 2017 dataset. The experimental results demonstrate the reliability and effectiveness of our suggested approach compared to existing methodologies. Our framework achieved a high level of accuracy (98.38%), precision (96.07%), recall (94.32%), dice score (95.07%), and Jaccard score (90.45%), outperforming current techniques.
ALL-FABNET: Acute Lymphocytic Leukemia Segmentation Using a Flipping Attention Block Decoder-Encoder Network Ammar S. Al-Zubaidi, Mohammed Al-Mukhtar, Ammar Awni Abbas Baghdadi Journal of Image and Graphics United Kingdom, 2024 Acute Lymphoblastic Leukemia (ALL) is a malignant neoplasm defined by the abnormal proliferation of immature lymphocytes in the hematopoietic system, specifically in the blood or bone marrow. The efficacy of ALL treatment is closely linked to its timely identification. Currently, the first diagnosis of ALL involves clinicians laboriously and fallibly examining stained blood smear microscopy images. Recently, deep learning techniques in biomedical diagnostics, focusing on human-centric approaches, have emerged as a potent tool to aid clinicians in their decision-making processes. As a result, researchers have devised a multitude of computer-aided diagnostic methods to detect ALL in blood images autonomously. However, most existing techniques for segmenting White Blood Cells (WBCs) do not consider the need for concurrent segmentation of the cytoplasm and nucleus. It is important to note that a significant drawback of the currently employed networks is their limited computational efficiency, which necessitates a substantial quantity of trainable parameters. The proposed deep learning model demonstrates favorable outcomes and can potentially be used to develop a dependable computer-aided detection system for leukemia malignancy. we suggest an Attention-Flipping Block (FAB) for the lightweight ALL-image segmentation model. It is evaluated using three publicly accessible datasets consisting of blood samples from individuals diagnosed with leukemia. These datasets are specifically referred to as C-NMC 2019, ALL_IDB1, and ALL_IDB2. With ALL-IDB2, the model’s segmentation accuracy is 93.56%, and its classification accuracy is 97.94 %, with an F1-Score of 97.65%.
Predicting COVID-19 in Iraq using Frequent Weighting for Polynomial Regression in Optimization Curve Fitting MD AL-Mukhtar, Ammar S. Al-Zubaidi, Mustafa N. Albadri Iraqi Journal of Science, 2024 The worldwide pandemic Coronavirus (Covid-19) is a new viral disease that spreads mostly through nasal discharge and saliva from the lips while coughing or sneezing. This highly infectious disease spreads quickly and can overwhelm healthcare systems if not controlled. However, the employment of machine learning algorithms to monitor analytical data has a substantial influence on the speed of decision-making in some government entities. ML algorithms trained on labeled patients’ symptoms cannot discriminate between diverse types of diseases such as COVID-19. Cough, fever, headache, sore throat, and shortness of breath were common symptoms of many bacterial and viral diseases. This research focused on the numerous tendencies and projected expansion of the Iraq pandemic to encourage people and governments to take preventive measures. This work is an established basic benchmark for demonstrating machine learning's capabilities for pandemic prediction. The suggested approach for forecasting the number of COVID-19 cases can assist governments in taking safeguards to avoid the disease's spread. We have demonstrated the effectiveness of our strategy using publicly available datasets and models. A polynomial network is trained on this premise, and the parameters are optimized using frequent weighting. When compared to linear models, the polynomial model predicts better and is more effective in forecasting COVID-19 new confirmed cases. As well, it aims to analyze the spread of COVID-19 in Iraq and optimize polynomial regression. In time series-based models, curve fitting using frequent weighting to implement models such as linear regression and polynomial regression is utilized to estimate the new daily infection number. The datasets were collected from March 13, 2020, to December 12, 2021. The continuous COVID-19 pandemic puts both human lives and the economy at risk. If AI could forecast the next daily hospitalization number, it may be a useful tool in combating this pandemic sickness.
GastroFPN: Advanced Deep Segmentation Model for Gastrointestinal Disease with Enhanced Feature Pyramid Network Decoder International Journal of Intelligent Engineering and Systems, 2024 The early identification of gastric cancer holds considerable importance within medicine due to its crucial role in mitigating fatality rates.Currently, artificial identification and annotations using gastroscopic images are the main methods of evaluation.Nevertheless, doctors have significant difficulties implementing these techniques due to the considerable heterogeneity in the visual characteristics of early cancer tumors.Weakness in segmentation remains the biggest obstacle to accurate detection and extraction of the main lesion from the tumor.In this paper, we propose a deep model combining two networks encoded: Unet++ and the feature pyramid network.The encoder backbone on first detection is based on ResNet34, which feeds the feature extraction to the next step.The second step is adding an enhanced feature pyramid network by merging blocks and final segmentation heads.The decoder improves the model's capacity to collect hierarchical characteristics at many levels, resulting in enhanced segmentation performance that adapts to changes in illness symptoms.The proposed model achieved a segmentation accuracy of 96.8%, a dice-score of 86.6%, and an F1-score of 85.3% when using the EDD2020 dataset.While the accuracy for DCSA-Unet achieved 92%, Unet++ 90%, 77% for FPN, and DeepLabv3+ 94.2%,We trained the proposed model on two different datasets, the CVC-ClinicDB and Kvasir-Seg datasets.For CVC-ClinicDB, the results metrics registered a Dice-Score 91.64%, an IoU of 84.63%, and an accuracy of 98.55%.For the Kvasir-Seg dataset, the Dice score is 92.54%, the IoU is 87.57%, and the accuracy is 96.62%.
Speech Enhancement Algorithm using Deep Learning and Hahn Polynomials Ammar S. Al-Zubaidi, Riyadh Bassil Abduljabbar, Basheera M Mahmmod, Sadiq H. Abdulhussain, Marwah A. Naser, Muntadher Alsabah, Abir Hussain, Dhiya Al-Jumeily Proceedings International Conference on Developments in Esystems Engineering Dese, 2024 Speech enhancement algorithms and machine learning can play a fundamental role in signal processing to improve speech quality. These techniques can be used to reduce noise and distortions in speech signals, hence ensuring clearer and more intelligible speech. By leveraging advanced machine learning, speech enhancement algorithms not only improve the listener’s auditory system, but also increase the efficacy of speech recognition systems. In particular, deep learning is a class of machine learning techniques, which have recently been used in speech enhancement. This paper proposes the use of Discrete Hahn polynomials (DHPs) o extract spectral features from noisy signals using fully connected neural networks and convolutional neural network. Deep learning can efficiently capture the contextual information of speech signals, resulting in superior improvements in speech quality and intelligibility properties. The results are evaluated based on the well-known TIMIT database. The results show that the presented model is able to enhance the speech signal for different conditions.
Low-Distortion MMSE Estimator for Speech Enhancement Based on Hahn Moments Ammar S. Al-Zubaidi, Basheera M. Mahmmod, S. Abdulhussain, M. Naser, Abir Hussain Proceedings International Conference on Developments in Esystems Engineering Dese, 2023 Discrete Hahn moments are considered efficient orthogonal moments applied in various scientific areas such as signal processing and computer vision. It has a high energy compaction, considered an advantage for speech enhancement algorithm (SEA). Most conventional SEA present undesirable distortion to the improved signal. Minimizing these issues demands a robust estimator. Therefore, this paper presents Hahn moments-based linear and non-linear estimators. Wiener filter and minimum mean squared error (MMSE) sense are used to form the estimators. These estimators with Hahn moments reduce the distortion in various underlying speech conditions. The presented SEA is evaluated in terms of different quality and intelligibility measurements. The experimental results show the advantage and effectiveness of the proposed system over other existing works.
