Medical image segmentation
Semantic segmentation for breast ultrasound, fetal ventricle delineation, and clinically relevant boundary-sensitive models.
Dhaka, Bangladesh
Medical Image Analysis
Semantic Segmentation
Production APIs
Researcher building clinically useful AI systems.
I am Arafat Hossain Sayem, a computer science graduate and software engineer working at the intersection of biomedical image analysis, segmentation models, and production-grade software systems. My current research focuses on robust and efficient deep learning methods for ultrasound and medical imaging workflows.
Current direction: I am seeking PhD opportunities and research collaborations in Biomedical AI, medical image segmentation, model robustness, machine learning and optimization.
Research Focus
Problems I care about
Efficient architectures
Lightweight CNN designs, MobileNet-based models, attention modules, and practical inference for limited-resource clinical settings.
Robustness and calibration
Understanding how segmentation systems behave under corruptions, domain shift, confidence errors, and real-world imaging variation.
Production software
Designing maintainable backend APIs using ASP.NET Core, CQRS, DDD, clean architecture, and scalable engineering practices.
Research
Manuscripts, preprints and ongoing work
Hybrid ResNet-50 Architecture for Multi-Dataset Breast Ultrasound Lesion Segmentation
Under review at Biomedical Signal Processing and Control (BSPC), Elsevier
Proposes a ResNet-50 based segmentation model with ASPP, Balanced CBAM attention, and anisotropic decoder convolutions for robust breast ultrasound lesion segmentation, with cross-dataset evaluation on unseen UDIAT data.
MVSegNet: A Lightweight Boundary-Aware Network for Fetal Lateral Ventricle Segmentation and Atrial Width Estimation in Prenatal Ultrasound
Submitted to ICCIT 2026
Presents a lightweight boundary-aware network for fetal lateral ventricle segmentation and atrial width estimation, targeting efficient prenatal ultrasound analysis in low-resource clinical settings.
Preventing Torn LoRA Updates in Live PEFT Serving
Submitted to ICECE, BUET, 2026
Identifies torn LoRA updates in live PEFT serving and introduces a lightweight double-buffered request-pinning mechanism so each request is served with one consistent adapter version without measurable added latency.
COD10K-C: Benchmarking Robustness of Camouflaged Object Detection Under Natural Image Corruptions
arXiv preprint, 2026
Introduces COD10K-C, a corruption robustness benchmark for camouflaged object detection under noise, blur, compression, rain, fog, brightness, and contrast corruptions, with evaluation across segmentation, boundary, and calibration metrics.
TaskRisk-Lite: Task-Aware Visual Reliability for Robot Manipulation
Ongoing research
Developing a lightweight model to predict unsafe robot actions caused by corrupted visual inputs using dual-camera vision, robot state, and task instructions on LIBERO-10.
Experience
Software engineering in production systems
Industry work presented as engineering maturity: architecture, APIs, maintainability, and user-facing systems.
Software Engineer
Bdjobs.com Ltd | Dhaka, Bangladesh
- Migrated core user-profile functionality from Classic ASP to ASP.NET Web APIs.
- Developed APIs for personal details, dashboard, photo upload, and job fair systems.
- Applied Domain-Driven Design, SOLID principles, CQRS, and event-driven architecture.
- Integrated backend APIs with Angular 18 frontend for production user workflows.
Trainee Software Engineer
Bdjobs.com Ltd | Dhaka, Bangladesh
- Implemented a core job repost API for the C2C platform.
- Built practical experience with Angular 18, Git/GitHub, and modern development workflows.
Capabilities
Research and engineering toolkit
Organized around the work you want to be known for: biomedical AI research plus maintainable software systems.
ML / Deep Learning
PyTorchCNNsResNetMobileNetU-NetCBAMASPPSemantic Segmentation
Medical Imaging
Breast UltrasoundPrenatal UltrasoundFetal Brain MRIBUSIBUS-UCLMUDIATData Curation
Research Tools
PythonNumPyOpenCVAlbumentationsMatplotlibGoogle ColabLaTeX
Software Engineering
ASP.NET Core 8C#Angular 18REST APIsCQRSDDDClean ArchitectureEntity Framework CoreJWTSQL
Education
Academic background
B.Sc. in Computer Science and Engineering
2020 - 2025
Stamford University Bangladesh | Dhaka, Bangladesh
Undergraduate thesis: Hybrid ResNet-50 Architecture for Multi-Dataset Breast Ultrasound Lesion Segmentation.
Contact
Open to PhD positions and research collaboration.
I welcome conversations with professors, research groups, and collaborators working on Biomedical AI, medical image analysis, robust segmentation, and trustworthy machine learning.