🔬 Research Interests
🩻 Medical Image Analysis
Developing deep learning models for disease detection and localization in chest X-rays and MRI, including tuberculosis detection and weakly supervised localization.
🧠 Explainable AI
Designing interpretable AI systems using techniques such as Grad-CAM, knowledge distillation, and centered kernel alignment (CKA) to make model decisions transparent for clinical use.
🖼️ Multimodal AI
Exploring multimodal approaches that fuse imaging data with clinical metadata to improve diagnostic performance and generalization across diverse patient populations.
👁️ Computer Vision
Applying convolutional neural networks, vision transformers, and classical image processing techniques to solve real-world problems in medical and dermoscopic imaging.
🔀 OOD Generalization
Studying how deep learning models generalize to out-of-distribution data, including the roles of weight initialization, data shuffling, loss landscape sharpness, and representation quality in robust generalization.
🎓 Education
Bangladesh University of Engineering and Technology (BUET)
Feb 2020 – Mar 2025
B.Sc. in Biomedical Engineering · Dhaka, Bangladesh
GPA: 3.83/4.00 (Rank 7th in class) · Relevant coursework: Medical Imaging, Digital Signal Processing, Linear Algebra, Probability & Statistics, Bioinformatics, Physiological Control Systems, Embedded Systems
💼 Experience
DSP
DSP Research Lab, BUET
Research Assistant
📌 Stroke risk prediction from PPG signals using ML; building a prototype to study the effect of external pressure on PPG waveform morphology.
Dec 2025 – Present · Dhaka, Bangladesh
mHL
mHealth Lab, BUET
Research Engineer
📌 Curated a large-scale dataset of 100,000 chest X-ray images from IEDCR; collaborated with radiologists to develop expert-annotated labels; drove pilot deployment of an in-house AI system at a local hospital.
May 2025 – Feb 2026 · Dhaka, Bangladesh
⚙️ Projects
Tuberculosis Detection from Chest X-ray Images
Nov 2023 – Mar 2025
Undergraduate Thesis · Supervisor: Dr. Taufiq Hasan, Professor, BME, BUET
- Trained CNNs (VGG, ResNet, DenseNet) and vision transformers (ViT, Masked Autoencoder) from scratch and via fine-tuning using PyTorch on a tuberculosis dataset.
- Visualized model attention with Grad-CAM and devised an algorithm to draw bounding boxes from heatmaps using OpenCV.
- Improved TB representation via knowledge distillation (soft labels) and analyzed the relationship between representation quality and OOD generalization.
- Computed centered kernel alignment (CKA) to measure inter-model similarity; analyzed loss landscape sharpness via largest eigenvalues.
- Explored the impact of randomness (weight initialization, batch shuffling) on out-of-distribution performance.
GPT-2 Pretraining from Scratch
Nov 2025
Personal Project
- Tokenized the BookCorpus dataset using Tiktoken and implemented the GPT-2 architecture in PyTorch.
- Utilized 4× NVIDIA GPUs on a cloud platform with PyTorch DistributedDataParallel (DDP) for scalable multi-GPU training.
- Applied modern training optimizations: mixed precision,
torch.compile, fused AdamW, Flash Attention, and gradient accumulation. - Built a custom DataLoader for efficient large-scale dataset handling.
Brain Tumor Classification from Magnetic Resonance Images
Feb 2023
PI: Shoyad Ibn Sabur Khan Nuhash, Assistant Professor, BME, BUET
- Implemented a custom model BrainMRNet from scratch incorporating attention modules, residual blocks, and hypercolumn techniques.
- Benchmarked BrainMRNet against VGG16, ResNet50, and DenseNet121 using accuracy, AUC, sensitivity, and specificity.
- Developed an interactive Streamlit web app for real-time local MRI inference.
ECG Signal Filtering and Heart Rate Measurement
Sep 2023
PI: Dr. Md. Kamrul Hasan, Professor, EEE, BUET
- Processed raw ECG signals from the MIT-BIH Arrhythmia Dataset and explored Butterworth, Chebyshev Type II, IIR Notch, and moving-average filters to remove noise.
- Applied wavelet transform for accurate QRS complex detection and heart rate estimation.
- Developed a MATLAB GUI application for ECG denoising and heart rate measurement.
Block-Based Hair Removal in Dermoscopic Images
Mar 2024
PI: Samiul Based Shuvo, Assistant Professor, BME, BUET
- Implemented a block-based algorithm for automatic hair removal in dermoscopic images using YIQ color space and morphological bottom-hat filtering.
- Performed block-wise inpainting via histogram-based pixel replacement followed by morphological closing to reconstruct lesion regions.
- Achieved performance comparable to the Dull Razor baseline across diverse skin tones and hair densities.
📝 Publications
🫁
Weakly Supervised Tuberculosis Localization in Chest X-rays through Knowledge Distillation
Tuberculosis (TB) remains one of the leading causes of mortality worldwide, particularly in resource-limited countries. Chest X-ray (CXR) imaging serves as an accessible and cost-effective diagnostic tool but requires expert interpretation, which is often unavailable. Although machine learning models have shown high performance in TB classification, they often depend on spurious correlations and fail to generalize. Besides, building large datasets featuring high-quality annotations for medical images demands substantial resources and input from domain specialists, and typically involves several annotators reaching agreement, which results in enormous financial and logistical expenses. This study repurposes knowledge distillation technique to train CNN models reducing spurious correlations and localize TB-related abnormalities without requiring bounding-box annotations. By leveraging a teacher-student framework with ResNet50 architecture, the proposed method trained on TBX11k dataset achieve impressive 0.2428 mIOU score. Experimental results further reveal that the student model consistently outperforms the teacher, underscoring improved robustness and potential for broader clinical deployment in diverse settings.
🛠️ Technical Skills
Programming Languages
ML Frameworks
Python Libraries
Software & Tools
Embedded Systems & Hardware
CAD & Simulations
🏆 Awards & Achievements
University Merit Scholarship
Bangladesh University of Engineering and Technology (BUET)
Regional Winner — Bangladesh Junior Science Olympiad
1st Prize — Regional Math Olympiad
🤝 Leadership & Activities
Vice President
Sep 2024 – May 2025
BUET Biomedical Engineering Society (BMES) Student Chapter
- Organized a nationwide scientific article writing competition engaging college and university students.
- Coordinated seminars and outreach programs on topics ranging from breast cancer awareness to student mental health.
Science Writer
Aug 2022 – Present
The Royal Scientific Publications Ltd
- Founded the monthly scientific magazine Divention, which reached over 10,000 high school and college students.
- Authored 40+ articles on topics including artificial intelligence and neuroscience to foster scientific curiosity among young readers.
Class Representative
Nov 2022 – Nov 2023
Bangladesh University of Engineering and Technology (BUET)
- Served as liaison between students and faculty; coordinated class schedules and examination dates.
- Advocated for student needs by resolving classroom infrastructure issues including sound systems and air conditioning.
😊 About Me
I am passionate about using AI to solve real-world problems in healthcare. Outside of research, I enjoy science writing and have founded Divention, a monthly magazine that brings science closer to young readers across Bangladesh. I am fluent in Bengali (native) and English (IELTS Band 8.0).
💬 I am open to research collaborations and opportunities. Feel free to reach out via email!