Md Moinul Azim

Intro

Hello! I’m Md Moinul Azim — a passionate data analyst and aspiring researcher with a strong background in machine learning, data visualization, and environmental/health data analysis.

I recently graduated in Computer Science and Engineering from the Bangladesh University of Engineering and Technology (BUET). My academic and project work reflects a commitment to using data for social impact — from predictive modeling in healthcare to visual analytics of global trends.

Welcome to my portfolio — a collection of carefully crafted projects that combine analytical depth with storytelling through tools like SQL, Python, Power BI, Tableau, and Excel.

Research

🎓 Undergraduate Thesis

Title: Comparative Study of U-Net Variants in QCT Bone Image Segmentation

Objective: To evaluate and compare the performance of different U-Net-based deep learning architectures—U-Net, Multi-Res U-Net, and Attention U-Net—for segmenting femur bones from Quantitative Computed Tomography (QCT) images.

Motivation: Accurate QCT image segmentation plays a crucial role in diagnosing bone diseases such as osteoporosis and in planning personalized treatments. Traditional manual segmentation is prone to inconsistency and expert-dependence. This study investigates deep learning models that can automate and improve segmentation accuracy, consistency, and efficiency.

Methodology:

• Preprocessed QCT images using normalization, voxel resampling, and augmentation.
• Implemented three deep learning architectures:
  • U-Net – A conventional encoder-decoder structure with skip connections.
  • Multi-Res U-Net – Introduces multi-resolution blocks to capture multi-scale features.
  • Attention U-Net – Employs attention gates to focus on relevant image regions.
• Compared models using metrics like Dice coefficient, Mean IoU, RMSE, MAE, and Relative Error for both surface area and volume segmentation.

Key Results:

• U-Net achieved the highest accuracy and the lowest error in both surface area and volume predictions.
• Multi-Res U-Net captured anatomical variations well but had slightly higher errors.
• Attention U-Net was more effective in identifying critical regions, though less accurate in surface area estimations.

Conclusion: The study concludes that while each variant has strengths, the standard U-Net model delivers the most balanced performance for femur segmentation in QCT images, especially in terms of reconstruction accuracy and error minimization​:contentReference[oaicite:0]{index=0}.

Tools & Technologies: Python, TensorFlow, NumPy, Medical Image Processing Libraries (e.g., SimpleITK, NiBabel), DICOM imaging.

Supervisor: Dr. Mahmuda Naznin, Department of CSE, BUET

📄 Read Full Thesis (PDF)

About Me

Hello! I’m Md Moinul Azim, a Computer Science and Engineering graduate from the Bangladesh University of Engineering and Technology (BUET) with a passion for data science, machine learning, and impactful research in healthcare and the environment.

My portfolio spans platforms including Python, SQL, Excel, Tableau, and Power BI. I’ve worked on diverse projects — from interactive dashboards analyzing COVID-19 and global layoffs to time series forecasting of environmental water quality and predictive modeling of heart disease risk.

For my undergraduate thesis, I conducted a comparative study of U-Net variants for segmenting femur bones from Quantitative Computed Tomography (QCT) images. I implemented and evaluated three architectures—U-Net, Multi-Res U-Net, and Attention U-Net—on clinical QCT data to determine their effectiveness in capturing anatomical structures. The study focused on segmentation accuracy and 3D reconstruction performance using metrics such as Dice coefficient, mean IoU, RMSE, and relative error. This work contributed insights for selecting optimal deep learning models in bone imaging tasks.

I’ve also contributed to public-facing analytical projects, including Power BI dashboards derived from international survey data, Excel-based dashboards for business insights, and predictive models for environmental and health outcomes.

Beyond academics, I’m deeply committed to teaching and mentoring — helping students prepare for university entrance and math Olympiads. I’ve received multiple national academic awards and have consistently earned a spot on the Dean’s List at the Bangladesh University of Engineering and Technology (BUET).

I’m driven by a goal to apply data and technology for social good. If you're interested in research collaboration, data-driven innovation, or impactful analytics — I’d love to connect!

Contact

• Email
• GitHub
• LinkedIn

Elements

i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;
    deck.shuffle();
    i++;
}

print 'It took ' + i + ' iterations to sort the deck.';