Boundary-Focused Benchmarking of Active Learning for Chest Radiograph Lung Segmentation Under Random Initialization
Benchmark-focused work on active learning with Dice, boundary Dice, and HD95 as central evaluation endpoints.
PhD Researcher
Medical AI, segmentation, and trustworthy evaluation.
I am Mohammed Mohaisen, a PhD researcher at the Artificial Intelligence Research Group, Budapest University of Technology and Economics, working on medical image segmentation, active learning, dataset shift, and boundary-aware evaluation.
About
Research profile
My research focuses on how medical AI systems behave under harder and more realistic conditions, including dataset shift, curation differences, annotation noise, and clinically meaningful boundary errors.
I am particularly interested in active learning for lung segmentation, HD95 and boundary Dice, and evaluation pipelines that make model claims more honest and reproducible.
Selected Work
Current manuscripts and research directions
A Guarded HD95-Aware Active-Learning Framework for Boundary-Critical Chest Radiograph Segmentation under Dataset Shift
Follow-on work on guarded acquisition, dataset shift, and cleaner adjudication of apparent improvements.
Mechanism analysis for hidden error and selection order
Ongoing work on why some acquisition strategies preserve useful difficulty signals while others lose them during realized selection.
Contact
Contact
Email: mohaisen@mit.bme.hu
Official page: BME PhD profile