Summer project
Bioinformatics Specialization · Cohort 2025
Deep Learning-Based Multi-Organelle Segmentation in Human Liver Electron Microscopy Images
Electron microscopy (EM) enables the high-resolution visualization of liver hepatocyte organelles; however, manual segmentation is time-consuming and requires specialized expertise. This project developed a deep learning pipeline using a 2D nnU-Net for multi-organelle segmentation of mitochondria, nuclei, endoplasmic reticulum (ER), lipid droplets, and cell boundaries, based on manually annotated datasets prepared with expert guidance. Evaluation across independent test images revealed organelle-specific performance: lipid droplets achieved the highest accuracy (Dice ≈ 0.81), nuclei moderate (≈ 0.70), mitochondria variable (≈ 0.62), and the ER the lowest (≈ 0.18), due to its structural complexity. These results demonstrate that limited high-quality annotations can effectively train deep learning models, establishing a scalable workflow for large-scale EM analysis and future 3D organelle modeling.
Mentor: Gary Bader
With KAUST Academy
- Cohort 2025
University of Toronto · Bioinformatics Summer Program
Want your entry updated or removed? Contact us.