The study titled "Maternal Health Risk Factors Dataset: Clinical Parameters and Insights from Rural Bangladesh" presents a comprehensive dataset of 1206 pregnant women from Kurigram General Hospital. It investigates the impact of pre-existing diabetes on maternal health risks using statistical tests including Chi-Square, Z-test, T-test, and ANOVA. Key findings reveal a strong association between pre-existing diabetes and higher pregnancy risk levels, significantly affecting BMI and other vital clinical indicators. The dataset serves as a critical resource for advancing maternal health research and developing predictive models in resource-constrained settings. 

My Contribution:

1. Developed and implemented data preprocessing and analysis code 

2. Performed Exploratory Data Analysis (EDA) for visual and statistical insights 

3. Conducted advanced statistical analysis using Chi-Square, Z-test, T-test, and ANOVA 

4. Contributed to writing the data article, summarizing findings and methodologies 

5. Ensured reproducibility and clarity of the analytical workflow 

The study "A Benchmark Dataset for Analyzing Hematological Responses to Dengue Fever in Bangladesh" presents a dataset of 1003 dengue patients from Kalai, Jaipurhat. It analyzes hematological parameters to identify gender-specific patterns and potential diagnostic markers. While no significant association was found between sex and overall diagnosis, statistically significant differences in hemoglobin levels between males and females were observed. The dataset holds promise for developing predictive models, improving clinical insights, and enhancing public health responses to dengue in endemic regions. 

My Contributions:

1. Implemented code for data preprocessing, cleaning, and normalization

2. Conducted Exploratory Data Analysis (EDA) and visualized key hematological trends

3. Performed advanced statistical analyses (Chi-Square, Z-test, T-test, ANOVA)

4. Co-authored the manuscript, contributing to data interpretation and insights

5. Designed reproducible analysis pipelines to support future research applications

The IEEE paper "Multi-Head Self-Attention Mechanisms in Vision Transformers for Retinal Image Classification" explores the application of Vision Transformers (ViTs) in the automated detection of Diabetic Retinopathy (DR). By leveraging multi-head self-attention mechanisms, the model effectively identifies critical features from retinal images. The approach achieved high performance metrics, including a 96.13% accuracy, 0.92 precision, 0.95 recall, 0.93 F1 score, and a 0.969 ROC-AUC score. This study highlights the potential of ViTs in improving diagnostic accuracy, reducing manual workload, and enabling earlier intervention in DR cases. Future enhancements aim to generalize the model across diverse datasets and integrate it into clinical workflows. 

My Contribution:

1. Designed and implemented the Vision Transformer (ViT) architecture with multi-head self-attention

2. Processed and preprocessed retinal image datasets for optimal model input

3. Conducted model training, evaluation, and hyperparameter tuning to maximize performance

4. Contributed to manuscript writing, experimental analysis, and interpretation of results

5. Ensured reproducibility of the pipeline for real-world medical deployment

The paper "An explainable AI-based blood cell classification using optimized convolutional neural network" presents a high-performing, interpretable deep learning approach for classifying four types of white blood cells (WBCs): eosinophils, lymphocytes, monocytes, and neutrophils. A novel, optimized CNN architecture achieved a testing accuracy of 99.12%, surpassing transfer learning baselines including Inception V3, MobileNetV2, and DenseNet201. To tackle the “black box” nature of deep learning, explainable AI (XAI) techniques such as SHAP, LIME, Grad-CAM, and Grad-CAM++ were integrated, enhancing model transparency. The model was successfully deployed in a user-friendly web and Android application, promoting real-world clinical usability. 

My Contribution:

1. Conceived the research idea and led the project from design to implementation

2. Designed and developed the optimized CNN architecture tailored for WBC classification

3. Applied and analyzed multiple Explainable AI (XAI) techniques (SHAP, LIME, Grad-CAM, Grad-CAM++)

4. Conducted comparative evaluation with state-of-the-art transfer learning models

5. Handled dataset curation, image pre-processing, model training, evaluation, and validation

6. Built and deployed the end-to-end system for web and Android platforms

7. Authored the manuscript and coordinated revisions with co-authors and reviewers

The paper “Tuberculosis Disease Detection from Chest X-rays Using Deep Learning Techniques” presents an efficient deep learning-based framework for identifying tuberculosis (TB) from chest X-ray (CXR) images. Leveraging a dataset of 3500 CXR images categorized into Tuberculosis and Normal classes, the study evaluates four prominent CNN architectures: VGG16, VGG19, MobileNetV2, and InceptionV3. Among these, MobileNetV2 outperformed the others, achieving 99.99% training accuracy and 98.93% test accuracy, making it the most robust model for automated TB detection. This research emphasizes the role of lightweight yet powerful models for real-world, scalable medical imaging applications.

My Contribution:

1. Lead the research direction and problem formulation focused on TB detection

2. Conducted dataset cleaning, augmentation, and image preprocessing

3. Implemented and fine-tuned all four CNN models

4. Analyzed model performances and identified MobileNetV2 as the best fit

5. Compiled results and visualizations, wrote the manuscript draft, and coordinated team inputs

This research presents a modified meander-shaped microstrip patch antenna designed specifically for Internet of Things (IoT) applications operating in the 2.4 GHz ISM band, widely used in commercial, research, and healthcare settings. The antenna is compact and optimized for integration with IoT devices, offering reliable performance with high gain, good efficiency, and small form factor, making it well-suited for short-range wireless communication.

My Contribution:

1. Design: Modified meander-line structure improves miniaturization and enhances current path length, leading to better impedance matching and radiation characteristics.

2. Simulation & Integration: Antenna was simulated and tested in conjunction with typical IoT modules and sensors for real-world applicability.

3. Performance Metrics: Ensures robust communication for wireless sensor networks, smart home systems, and healthcare monitoring.