nnQC: Neural Network Quality Control for Medical Image Segmentation

A comprehensive toolkit for training, evaluating, and improving medical image segmentation models using diffusion-based approaches with cross-attention mechanisms.

🎯 Overview

nnQC provides tools for:

• Training autoencoder models for medical image latent representations
• Training diffusion models with cross-attention for segmentation refinement
• Evaluating model performance with comprehensive metrics
• Generating synthetic segmentation masks for quality control

🏗️ Architecture

The system consists of three main components:

1. Autoencoder: Encodes medical images and segmentation masks into latent space
2. Diffusion Model: Generates refined segmentation masks using cross-attention with CLIP features
3. Evaluation Pipeline: Comprehensive metrics computation including DSC, HD95, and correlation analysis

📦 Installation

From Source

git clone https://github.com/yourusername/nnQC.git
cd nnQC
pip install -e .

Requirements

• Python >= 3.8
• PyTorch >= 1.12.0
• MONAI >= 1.2.0
• CUDA-capable GPU (recommended)

See requirements.txt for complete dependency list.

🚀 Quick Start

Using as a Package

import nnqc

# Train autoencoder
nnqc.train_autoencoder()

# Train diffusion model  
nnqc.train_diffusion()

# Run evaluation
results = nnqc.evaluate_validation_set(args)

Command Line Interface

# Train autoencoder
nnqc-train-ae -c config/config_train_32g.json -g 2

# Train diffusion model
nnqc-train-diffusion -c config/config_train_32g.json -g 2

# Run inference/evaluation
nnqc-inference -c config/config_train_32g.json

# Evaluate validation set
nnqc-evaluate -c config/config_train_32g.json

Alternative CLI (using module syntax)

# Train autoencoder
python -m nnqc.training.train_autoencoder -c config/config_train_32g.json

# Train diffusion  
python -m nnqc.training.train_diffusion -c config/config_train_32g.json

# Run inference
python -m nnqc.inference.inference -c config/config_train_32g.json

📁 Project Structure