PROCESS CONTROL, IMAGE PROCESSING, POWER ELECTRONICS AND INTELLIGENT CONTROL
62
Scopus Publications
Scopus Publications
Brain MRI Analysis Using SqueezeNet for Accurate Medical Image Classification Dhanalakshmi S, Arulselvi S Proceedings of the 2025 11th International Conference on Communication and Signal Processing Iccsp 2025, 2025 Early diagnosis and efficient treatment of neurological illnesses need accurate brain Magnetic Resonance Imaging (MRI) classification. The suggested approach improves brain MRI analysis using SqueezeNet, a lightweight convolutional neural network. The goal is to achieve high accuracy with low computing complexity to meet medical imaging's rising need for scalable solutions. Improve classification performance, reduce false predictions, and adapt to different datasets. Its small design optimizes parameter utilization and accuracy, making SqueezeNet suited for resource-constrained systems. Fire modules and global pooling layers balance computational efficiency and classification accuracy in the model. Performance indicators like accuracy, sensitivity, and specificity reveal the model's robustness. A dependable and scalable solution to help healthcare practitioners quickly and accurately detect brain MRI abnormalities is the aim. This study advances medical imaging technology, especially in resource-constrained settings, by allowing accurate diagnosis. To improve SqueezeNet's performance and generalizability, transfer learning and data supplementation will be explored. The REMBRANDT database shows that SqueezeNet v1.2 performs best. The SqueezeNet v1.2 approach offers 97.4% sensitivity, 96.5% specificity, and 96.9% accuracy.
Brain MRI Analysis with Ensemble Classification Methods for Enhanced Diagnostic Precision Dhanalakshmi S, Arulselvi S Proceedings of the 2025 11th International Conference on Communication and Signal Processing Iccsp 2025, 2025 The objective of enhancing Brain MRI analysis with ensemble classification approaches is to improve diagnosis accuracy for neurological disorders. The goal is to amalgamate many machine learning algorithms to create a resilient classification system that capitalizes on the advantages of each model while alleviating their shortcomings. Ensemble approaches, by integrating techniques like random forests, support vector machines, and neural networks, may provide a more precise and dependable diagnosis of brain problems, including tumors and neurodegenerative diseases. The objective is to attain enhanced sensitivity and specificity, reducing false positives and false negatives, so empowering doctors to make more informed judgements. This method aims to decrease analysis duration while enhancing diagnostic consistency, making it appropriate for practical clinical use. Ensemble classification approaches in Brain MRI analysis signify a significant leap in medical imaging and artificial intelligence within healthcare. Results from the REMBRANDT database indicate that Bagging in Ensemble Classification yields the most favorable outcome. The Bagging method has a sensitivity of 99.6%, specificity of 99.8%, accuracy of 99.7%.
Deep Learning-Enhanced PET-MRI Fusion for Automated Brain Tumor Detection and Classification N. Aashna Unnikrishnan, S. Arulselvi, B. Karthik Proceedings of 3rd IEEE International Conference on Knowledge Engineering and Communication Systems Ickecs 2025, 2025 The study proposes a DL-assisted fusion model of PET-MRI scans for the automation of brain tumor identification and classification to get around the aforementioned problems. Existing techniques cannot be completely precise and quick since these only use one of these modalities (MRI or PET) to examine tumor attributes. Through a multi-input CNN-Based- Attention mechanism, the proposed technique combines structural MRI and functional PET data, allowing for the effective aggregation of complementing information from both modalities to provide a thorough picture of the tumor. With high diagnostic precision, the automatic feature extraction, fusion, and classification reduces processing time and human error. With 96.8% accuracy, 94.5% sensitivity, and 98.2% specificity, the results exhibit notable improvements over existing systems and significantly outperform earlier methods in terms of tumor classification and lower false positive/negative rates. The proposed technique has the potential to be a viable clinical tool for brain tumor diagnosis due to its enhanced categorization precision for benign, malignant, and abnormal tissues.
AI-Driven Denoising and Image Fusion Techniques for PET-Based Brain Tumor Detection N. Aashna Unnikrishnan, S. Arulselvi, B. Karthik Proceedings of 3rd IEEE International Conference on Knowledge Engineering and Communication Systems Ickecs 2025, 2025 Brain tumors must be detected optimally to allow correct identification and treatment. PET and MRI scans are widely used in brain imaging, while PET scans are often noisy which hampers accurate detection. An AI-based model thatuses U-Net to learn from a denoising CNN for image, multi-modal image fusion, and tumor segmentation. The proposed point denoising model also reached better denoising compared with previous solutions, with the achieved PSNR of 37.2 dB and SSIM of 0.91. The segmentation module achieved a high DSC of 0.89, similar to previous models based on fusion. The framework was also validated on the ADNI PET-MRI dataset for efficacy improvement over previous work achieving a PSNR improvement of $10. \%$ & DSC improvement of 13. %. From these outcomes, innovation’s possibility for improved detection of tumors through the system has been emphasized thus making its clinical application more reliable.
Hybrid Feature Fusion Approach for Precise Brain Tumor Detection and Grading Using Vision Transformers N. Aashna Unnikrishnan, S. Arulselvi, B. Karthik Proceedings Iceconf 2025 2025 2nd International Conference on Artificial Intelligence and Knowledge Discovery in Concurrent Engineering, 2025 Accurate diagnosis and grading of brain tumors is critical for appropriate clinical intervention; however, previous tumor detection systems based on handmade features or CNN-only frequently fail to identify complex tumor characteristics, resulting in poor generalization and robustness. Existing systems struggle to model the global context, are unable to accommodate noise reliably, and deliver low overall classification accuracy. To address the aforementioned issues, it presents a hybrid feature fusion system that extracts local spatial features using CNN s while modeling global context with Vision Transformers. All of these fused characteristics were run through an attention-guided pipeline to improve tumor region discrimination. On a test dataset of 100 patients with annotated annotations, the proposed system achieved higher binary classification accuracy (96.4%), precision (95.7%), and AUC (97.8%) for MRI datasets, as well as multi-class grading accuracy of 91.4%. The model's reliability, interpretability, and practicality in clinical contexts outperform traditional techniques, as evidenced by these results.
Integration of Generative AI and Image Fusion for Enhanced Brain Tumor Detection in PET-MRI Data N. Aashna Unnikrishnan, S. Arulselvi, B. Karthik 3rd IEEE International Conference on Device Intelligence Computing and Communication Technologies Dicct 2025, 2025 This study introduces a novel generative AI-driven system for brain tumor detection by integrating PET and MRI data through an advanced image fusion approach. The fused images were used for training on the machine in the real-world dataset. It provided very high sensitivity and specificity with a sensitivity of 98.7 and a specificity of 99.2%. Ablation experiments with statistical analysis demonstrated the superiority of the system along with its robustness with other existing models. The viability of the technology for healthcare applications was also illustrated through real-time implementation through an intuitive interface. This research introduces a novel approach to diagnosing brain malignancies using multimodal imaging to set new benchmarks for accuracy and reliability.
Visual Geometry Group Architectures for Brain Cancer Diagnosis using MRI Scans S. Dhanalakshmi, S. Arulselvi 4th International Conference on Innovative Practices in Technology and Management 2024 Iciptm 2024, 2024 The Visual Geometry Group Architecture Technique (VGGAT) is a powerful deep learning brain tumor classification system that uses MRI images. This research work analyzes its ability to effectively diagnose brain scan malignancies. In order to learn complicated features from raw data, VGGAT makes use of Convolutional Neural Network (CNN) architecture. VGGAT acquires the ability to distinguish between healthy tissue and malignant tissue with a high degree of precision via the process of training on a huge dataset of annotated MRI images. Using thorough assessment on separate test sets, VGGAT displays strong performance in classification tasks, exceeding standard approaches. This is accomplished via the analysis of data. The purpose is to emphasize the usefulness of VGGAT in supporting doctors with accurate and fast diagnosis of brain cancer, which in turn facilitates early intervention and better patient outcomes. Results proved that the proposed VGG system has achieved with 94.29% overall accuracy with 94.2% sensitivity and 94.4% specificity.
AlexNet Architecture for Classifying of Magnetic Resonance Imaging Scans of the Brain S. Dhanalakshmi, S. Arulselvi 2024 International Conference on Advances in Modern Age Technologies for Health and Engineering Science Amathe 2024, 2024 Classification of MRI brain images using AlexNet and deep learning algorithms is the goal of this study. The goal is to create a reliable and effective system that can classify Magnetic Resonance Imaging (MRI) scans into meaningful clinical categories, such as normal, benign, or cancerous. Improving the diagnosis of neurological illnesses and anomalies is the primary objective of using deep learning, namely the AlexNet architecture. This work aims to enhance medical decision-making and patient care by developing and validating a system that can reliably categorize brain images to high levels of accuracy. The end goal is to provide a flexible and extensible framework that can be included into existing clinical procedures; this will help medical personnel make faster diagnoses based on better MRI scan interpretation. Improving neuroimaging diagnostic tools and procedures is the ultimate goal of this study, which aims to expand the frontiers of medical image analysis. Results proved that the proposed AlexNet system achieves 95.0 % overall accuracy with 93.8% sensitivity and 99% specificity.
Analysing Osteoporosis Detection: A Comparative Study of CNN and FNN R. Geetha, S. Arulselvi, R. Tamilselvi, M.Parisa Beham, Alavikunhu Panthakkan, Wathiq Mansoor, Hussain Al Ahmad 2024 7th International Conference on Signal Processing and Information Security Icspis 2024, 2024 Osteoporosis causes progressive loss of bone density and strength, causing a more elevated risk of fracture than in normal healthy bones. It is estimated that some 1 in 3 women and 1 in 5 men over the age of 50 will experience osteoporotic fractures, which poses osteoporosis as an important public health problem worldwide. The basis of diagnosis is based on Bone Mineral Density (BMD) tests, with Dual-energy X-ray Absorptiometry (DEXA) being the most common. A Tscore of - 2.5 or lower defines osteoporosis. This paper focuses on the application of medical imaging analytics towards the detection of osteoporosis by conducting a comparative study of the efficiency of CNN and FNN in DEXA image analytics. Both models are very promising, although, at $95 \%$, the FNN marginally outperformed the CNN at $93 \%$. Hence, this research underlines the probable capability of deep learning techniques in improving the detection of osteoporosis and optimizing diagnostic tools in order to achieve better patient outcomes.
Technical advancement and social challenges associated with functional capabilities of 5G cellular technologies Department of Electronics, Communication Engineering, Bharath Institute of Higher Education, Research, Chennai, Tamilnadu, India., S. Arul Selvi, S. Saravana, Department of Electronics, Communication Engineering, Bharath Institute of Higher Education, Research, Chennai, Tamilnadu, India., G. Kanagavalli, et al. International Journal of Engineering and Advanced Technology, 2019
Neighbor discovery in ASN M. Jasmin, S. Philomina, S. Arulselvi, Jesse Liberty, Dan Maharry, et al. International Journal of Innovative Technology and Exploring Engineering, 2019
Multifactor optimized clustering with improved scheduling for receiver-initiated mac International Journal of Engineering and Advanced Technology, 2019
An effective intra and inter cluster formation with scheduling technique for a WSN system International Journal of Innovative Technology and Exploring Engineering, 2019
Receiver-initiated medium access control (RI-MAC) protocols International Journal of Innovative Technology and Exploring Engineering, 2019
Construction of ensemble square classification approaches in MIMO OFDM International Journal of Engineering and Advanced Technology, 2019
ES-MAC: A sink-aware beacon scheduling transmission for receiver-initiated mac protocol for wireless sensor network International Journal of Recent Technology and Engineering, 2019
IOT based industrial automation International Journal of Recent Technology and Engineering, 2019
Optimized method of spectrum sensing in cognitive radio networks International Journal of Recent Technology and Engineering, 2019
Area and power budget estimation of hierarchical network topology in comparison with 2D mesh topology for NOCs and its design and implementation oriented overview International Journal of Recent Technology and Engineering, 2019
A stochastic analysis on translating Nam speech into normal speech International Journal of Recent Technology and Engineering, 2019
Multi-core micro-controller architecture with ZLPIC for high performance embedded applications International Journal of Innovative Technology and Exploring Engineering, 2019
Granular traffic analysis and energy modeling in NoC with enhanced data transmission International Journal of Innovative Technology and Exploring Engineering, 2019
Designing network interface component for peripheral IP cores in networks-on-chip International Journal of Innovative Technology and Exploring Engineering, 2019
A combined framework for routing and channel allocation for dynamic spectrum sharing using cognitive radio International Journal of Applied Engineering Research, 2016
A new intelligent human walking cane type robot International Journal of Applied Engineering Research, 2014
Novel anti theft system using Zigbee –pic remote International Journal of Applied Engineering Research, 2014
Low power VIsi simulation of advanced encryption standard substitution box against side channel attack International Journal of Applied Engineering Research, 2014
Fault protected encoder and decoder for nano memory applications International Journal of Applied Engineering Research, 2014
