Yijian L.

Key Skills

Software

Don't disclose

Other

Top Subject Matter

Medical image segmentation (brain tumors, MRI modalities)

Unsupervised/automatic video object segmentation

Pressure-sensitive sensing and weak-signal processing for object detection

Top Data Types

Video

Image

Top Task Types

Segmentation

Tracking

Transcription

Freelancer Overview

Brain Tumor Detection Based on Deep Learning (Project Lead). Brings 1+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Don't disclose and Other. Education includes Master of Science, Nanyang Technological University (2026) and Bachelor of Science, University of Central Lancashire (2023). AI-training focus includes data types such as Medical, DICOM, and Video and labeling workflows including Segmentation and Tracking.

Labeling Experience

Unsupervised Video Object Segmentation (UVOS) (Project Lead)

Don't discloseVideoTracking

Conducted an unsupervised video object segmentation study on YouTube-VOS, implementing and comparing inference pipelines that require labeled evaluation targets and structured ground-truth assessment. Designed training and evaluation workflows using optical-flow pseudo-labels and SAM+XMem hybrid inference to segment moving objects through frames. Assessed model performance with quantitative metrics and success/failure case analysis. • Implemented an optical-flow plus U-Net pipeline with pseudo-label generation • Built a SAM plus XMem hybrid inference pipeline for video segmentation • Evaluated with loss functions, PR curves, IoU, Precision, and Recall • Performed temporal stability and complex-scene adaptability comparisons

2025 - Present

Brain Tumor Detection Based on Deep Learning (Project Lead)

Don't discloseSegmentation

Led a multimodal medical image segmentation project using the BraTS 2023 PED brain tumor MRI dataset, focusing on building and validating labeled training/evaluation data workflows for segmentation. Preprocessed multiple MRI modalities into multichannel inputs and trained a 2D U-Net style encoder-decoder model to produce pixel-wise tumor labels. Evaluated the segmentation with standard metrics and analyzed performance differences across tumor subregions and modalities. • Used Dice, IoU, Precision, Recall, and F1 to quantify segmentation quality • Visualized results and compared performance across MRI modalities (e.g., T1C, T1N, T2F/FLAIR, T2W) • Documented findings on modality contributions to lesion-range and boundary representation • Prepared analysis for multimodal fusion and lesion localization strengths/limitations

2025 - Present

Education

Work History