This repository contains an implementation of HDC-Net, a deep learning architecture designed for accurate segmentation of 3D medical volumes, such as CT or MRI scans. It is especially effective for liver tumor segmentation, but the framework is adaptable to other 3D biomedical imaging tasks.
HDC-Net (Hybrid Dilated Convolutional Network) combines:
- U-Net structure for encoder-decoder learning
- Hybrid Dilated Convolutions for better multi-scale feature extraction
- Residual and skip connections to maintain spatial precision
- 🚀 Enhanced receptive field with hybrid dilated convolutions
- 🧱 Modular and easy-to-read 3D implementation
- 🧪 Trained on volumetric data from liver segmentation datasets
- 🎯 Supports patch-wise training for low-memory environments
- 🧾 Checkpointing and prediction saving
Trained on https://www.kaggle.com/datasets/andrewmvd/liver-tumor-segmentation