tbhswanand/brats-mri-segmentation

🧠 BraTS MRI Tumor Segmentation (UNet-based)

📌 Overview

This model performs brain tumor segmentation on MRI scans using a deep learning architecture (UNet-based). It is trained on the BraTS dataset for segmenting tumor regions such as:

• Enhancing Tumor (ET)
• Tumor Core (TC)
• Whole Tumor (WT)

---

🧪 Model Details

• Architecture: UNet / UNet-like
• Framework: PyTorch
• Input: MRI slices (multi-modal or single modality)
• Output: Segmentation mask
• Task: Medical Image Segmentation

---

📂 Dataset

• Dataset: BraTS (Brain Tumor Segmentation Challenge)

• Modalities: T1, T1c, T2, FLAIR

• Preprocessing:

  • Normalization
  • Resizing
  • Skull stripping (if applied)

---

⚙️ Usage

🔹 Install dependencies

pip install torch torchvision nibabel numpy

---

🔹 Load model

import torch

model = torch.load("model.pth", map_location="cpu")
model.eval()

---

🔹 Run inference

import numpy as np

# Example input (replace with actual preprocessing)
input_tensor = torch.randn(1, 4, 240, 240)

with torch.no_grad():
    output = model(input_tensor)

mask = output.argmax(dim=1)

---

📊 Performance

Metric	Score
Dice Score (WT)	~0.85
Dice Score (TC)	~0.80
Dice Score (ET)	~0.75

Note: Values may vary based on training setup

---

🚀 Applications

• Clinical decision support
• Tumor monitoring
• Radiology AI systems
• Medical research

---

⚠️ Limitations

• Not for direct clinical use
• Performance depends on preprocessing
• Requires GPU for fast inference

---

🔐 License

Specify your license here (MIT / Apache 2.0 / etc.)

---

👨‍💻 Author

Swanand Wakadmane AI/ML Developer | Medical AI Enthusiast

---

🤝 Acknowledgements

• BraTS Challenge
• Medical imaging research community