Graduate project/demo: AI-Based Motion Artifact Correction and Active Learning Framework for Enhanced Brain Tumor Segmentation
SegmentationDeveloped an AI-based motion artifact correction and active learning framework to enhance brain tumor segmentation and reduce annotation cost. Designed a labeling-efficient training pipeline using uncertainty estimation to select the most informative samples for human annotation. Evaluated segmentation improvements using quantitative image quality metrics on a new real-world dataset.• Achieved state-of-the-art segmentation-related improvements with motion correction, reporting SSIM and PSNR gains.• Built an uncertainty-driven active learning pipeline to optimize model performance with minimal additional annotations.• Collaborated in a clinical-like setting for pediatric GBM segmentation, integrating AI into a cloud-based medical platform.• Utilized deep learning frameworks and medical imaging toolchains to train and test the system.