Skip to content

code_availability_statement.md

Latest commit

 

History

History
91 lines (65 loc) · 2.65 KB

code_availability_statement.md

File metadata and controls

91 lines (65 loc) · 2.65 KB

Code Availability Statement

Overview

The complete source code for AFN-DeSeg (Auto-Fluorescence Nuclei Denoising and Segmentation) is publicly available to ensure full reproducibility of the results presented in this study.

Repository

GitHub Repository: https://github.com/z-pan/AFN-DeSeg

Contents

The repository includes:

  1. Model Implementation

    • AFN-DeSeg dual-encoder architecture (U-Net + DINOv3 ViT)
    • LoRA adaptation modules for efficient fine-tuning
    • Attention U-Net for Key Diagnostic Area (KDA) prediction

  2. Training Scripts

    • Stage 1: DINO domain adaptation for TPAF microscopy
    • Stage 2: Joint denoising and segmentation training
    • KDA model training

  3. Data Processing

    • Mixed Poisson-Gaussian (MPG) noise synthesis
    • Data augmentation pipelines
    • Dataset loaders for TPAF images

  4. Evaluation

    • Comprehensive metrics (PSNR, SSIM, Dice, IoU, mAP, HD95)
    • KDA-specific metrics (KAF, nuclear density correlation)
    • Bland-Altman analysis for clinical validation

  5. Inference

    • Prediction scripts for new TPAF images
    • Batch processing utilities

System Requirements

Software Dependencies

  • Python >= 3.8
  • PyTorch >= 2.0.0
  • CUDA >= 11.7 (for GPU acceleration)

Hardware Requirements

  • GPU: NVIDIA GPU with >= 8GB VRAM (recommended: RTX 3090 or A100)
  • RAM: >= 32GB system memory
  • Storage: >= 10GB for code, models, and sample data

Typical Installation Time

  • Fresh installation: ~10-15 minutes
  • With pre-downloaded model weights: ~5 minutes

Reproducibility

Pretrained Weights

Pretrained model weights are available at:

  • DINOv3 backbone: facebook/dinov3-vitb16-pretrain-lvd1689m (HuggingFace)
  • AFN-DeSeg checkpoint: Available in the checkpoint/ directory

Random Seeds

All experiments use fixed random seeds for reproducibility:

  • PyTorch seed: 42
  • NumPy seed: 42
  • CUDA deterministic mode: enabled

Expected Runtime

  • Training (Stage 2, 150 epochs): ~12-24 hours on single A100 GPU
  • Inference (per 512×512 image): ~50-100ms on GPU

License

This code is released under the MIT License, allowing free use, modification, and distribution for both academic and commercial purposes with proper attribution.

Citation

If you use this code in your research, please cite:

Pan, Z., Song, N., Cheng, S., et al. A Joint Denoising and Segmentation Framework
for Ovarian Cancer Diagnosis based on Two-Photon Autofluorescence Microscopy.
Nature Communications (2024).

Contact

For questions regarding the code implementation: