Medical Image Analysis: LV Segmentation

Workshop on Cardiac MRI segmentation using U-Net architecture. Covering DICOM processing and Dice-based loss functions.

This specialized workshop focuses on Medical Image Analysis, specifically the segmentation of the Left Ventricle (LV) from Cardiac MRI scans. Using the U-Net architecture, we explore how deep learning can assist in automated cardiac diagnostics.

Workshop Instructor: Zahra Amini

The repository includes the complete pipeline: from handling medical imaging formats to implementing custom loss functions and evaluating the model’s performance on clinical datasets.

🚀 ACCESS WORKSHOP MATERIALS ON GITHUB


Workshop Highlights

The curriculum is designed to cover the end-to-end medical segmentation workflow:

  • Medical Data Handling: Processing DICOM/NIfTI files and data augmentation techniques specifically for MRI.
  • Architecture Deep-Dive: Understanding the symmetric encoder-decoder structure of U-Net and skip connections.
  • Loss Functions for Imbalance: Implementing Dice Loss and Intersection over Union (IoU) to handle class imbalance in medical masks.
  • Evaluation: Analyzing segmentation masks and measuring pixel-wise accuracy in a clinical context.

Technologies used: PyTorch/TensorFlow, OpenCV, Nibabel, Scikit-Image.