Overview

This repository provides a workflow for medical image segmentation from CT scans.
For each organ, we run Two AI models (TotalSegmentator, and ST-UNet), visualize results in 2D and 3D, and compute evaluation metrics: IoU, Dice, ASSD (mm).
A lightweight GUI lets you switch organs/models and control color, visibility, opacity, and run evaluation interactively.

---

Key Features

• Modular per-organ workflow: plug in multiple models per organ
• 2D visualization: random slice preview with mask overlay for quick sanity checks
• 3D visualization: smooth meshes per organ (often 3+ parts with distinct colors)
• Built-in metrics: IoU, Dice, ASSD (mm)
• Interactive GUI: choose organ and model, adjust color, visibility, opacity, and Run Evaluation
• Reproducible scripts: consistent steps for download/setup/run

---

Pipeline Steps

1. Dataset download (public CT datasets from the internet)
2. Code integration & model wiring per organ
3. 2D visualization of a random CT slice with predicted mask overlay
4. 3D visualization with multi-part organ meshes (3+ segments, distinct colors)
5. Evaluation with IoU, Dice, ASSD (mm)
6. GUI to switch organs/models and run evaluation on demand

---

Setup Me "powershell"

pip install numpy scipy matplotlib scikit-image nibabel

pip install pyqt5 pyvista pyvistaqt

pip install torch torchvision torchaudio

pip install medim

pip install TotalSegmentator

---

Models

• TotalSegmentator — whole-body labels (104 classes) re-used per organ
• ST-UNet — U-Net variant adapted for CT volumes

Each organ can map to any subset of these models; results are saved per model to enable side-by-side comparison.

---

Evaluation Metrics

• Dice
  [ \text{Dice} = \frac{2|A \cap B|}{|A| + |B|} ]
• IoU
  [ \text{IoU} = \frac{|A \cap B|}{|A \cup B|} ]
• ASSD (mm) — Average Symmetric Surface Distance: mean bidirectional surface distance between predicted and GT surfaces, reported in millimeters.

---

Licenses

Model / Framework	License	Notes / Source
TotalSegmentator	Apache License 2.0 ([GitHub][1])	The official repo states “Apache-2.0 license” ([GitHub][1])
[1]: https://github.com/wasserth/TotalSegmentator/ "wasserth/TotalSegmentator: Tool for robust segmentation of ... - GitHub"

---

Team & Credits

Team Members

• Abdallah Saeed
• Bassel Mostafa
• Mahmoud Zahran
• Rawan Kotb

Supervised by

• Prof. Tamer Basha
• Eng. Alaa Tarek