🧠 3D U-Net for Medical Image Segmentation

This repository contains a clean and customizable implementation of a 3D U-Net model, designed for volumetric (3D) medical image segmentation tasks. It is particularly useful for segmenting organs or tumors in CT or MRI scans.

---

🏥 Application

This implementation was tested on the Liver Tumor Segmentation (LiTS) dataset. It can be adapted for any medical image dataset that provides 3D volumetric data (NIfTI or NumPy format).

---

📌 Features

• ✅ 3D U-Net architecture with encoder-decoder structure
• ✅ Works with NIfTI or .npy 3D data
• ✅ Patch-based training for memory efficiency
• ✅ Visualizes slices and segmentation results
• ✅ Easy integration with new datasets
• ✅ Save/load model checkpoints

---

🧱 Model Architecture

The 3D U-Net consists of:

• Encoder: 3D Convolution → BatchNorm → ReLU → MaxPool
• Bottleneck: Deepest feature representation
• Decoder: Transposed 3D Convolutions + Skip Connections
• Output: 1×1×1 convolution for voxel-wise classification

---

Trained on https://www.kaggle.com/datasets/andrewmvd/liver-tumor-segmentation