Important

This repository is the official implementation of Adaptive-SN2N.

Adaptive-SN2N

Artifact-suppressed and adaptive self-inspired learning denoising for super-resolution fluorescence microscopy.

This repository contains the official source code for our paper, "Artifact-suppressed and adaptive self-inspired learning denoising for super-resolution fluorescence microscopy". This work introduces Adaptive-SN2N, an enhanced self-inspired learning framework for image denoising in fluorescence microscopy. Our method is specifically designed to suppress background artifacts, a common challenge in biological image analysis, by incorporating a risk-aware adaptive normalization strategy and a Gaussian-weighted overlap-tile inference mechanism.

Published paper DOI: https://doi.org/10.3724/PXLIFE.2025-0010

✨ Introduction | 🔧 Installation | 🎨 Data Generation | 💻 Training & Inference | 📜 License

---

✨ Introduction

Adaptive-SN2N is an enhanced self-inspired learning framework for image denoising in fluorescence microscopy. It overcomes the limitations of standard self-supervised methods by:

1. Suppressing Background Artifacts: Utilizing a risk-aware adaptive normalization strategy.
2. Adaptive Inference: Implementing a Gaussian-weighted overlap-tile inference mechanism for seamless reconstruction.
3. Dual-Mode Learning: Supporting both global and local learning modes to adapt to different noise characteristics.

🔧 Installation

Dependencies

• Python >= 3.8
• PyTorch >= 1.12
• Other dependencies: numpy, scipy, scikit-image, tifffile, matplotlib, pandas

Instruction

1. Clone the repository.

   git clone https://github.com/YourUsername/aSN2N.git
   cd aSN2N    

2. Install the required dependencies.

   pip install -r requirements.txt

🎨 Data Generation

Before training the model, you need to generate an unsupervised training dataset from your raw microscopy images(preferably with images in tif format) using Scripts_aSN2N_datagen.py. This script generates "Global" and "Local" mode datasets required for the adaptive learning process.

Usage

Run the script from the command line:

python Scripts_aSN2N_datagen.py --train_data_path "path/to/raw/images" --output_base_path "path/to/save/data" --both_modes

Parameters

    -----Parameters------
    --train_data_path: (Required)
        Path to the directory containing your raw training images (e.g., .tif files).
    --output_base_path: (Optional, default: './output')
        Base path where the generated datasets will be saved. 
        Subdirectories 'global' and 'local' will be created automatically.
    --both_modes: (Optional, flag)
        If set, generates BOTH global and local datasets regardless of the adaptive decision.
    --vis_patches: (Optional, flag)
        Enable visualization of individual patches for debugging/analysis.
    --vis_overlay: (Optional, flag)
        Enable visualization of risk overlay on images.
    --export_csv: (Optional, flag)
        Export patch metrics to CSV for analysis.

💻 Training & Inference

The training and inference processes are integrated into Scripts_aSN2N_train.py. This script supports multi-GPU training and is configured via a JSON file.

1. Configuration

Create a configuration file (e.g., Config/your_config.json) based on the provided example in Config/example.json.

Example Configuration (Config/example.json):

[
    {
        "dataset_name": "experiment_name",
        "train_data_path": "path/to/generated/training_data",
        "test_path": "path/to/raw/data/for/inference",
        "epochs": 60,
        "train_batch_size": 32,
        "test_batch_size": 1,
        "reg_sparse": 0,
        "reg": 0.5,
        "work_mode": "local",
        "inference_mode": "local"
    }
]

• work_mode: Set to "local" or "global" depending on the dataset being trained.
• inference_mode: Defines the inference strategy (usually matches work_mode).

2. Execution

Before running the script, ensure the config_path in Scripts_aSN2N_train.py points to your configuration file, or modify the script to load your specific JSON file.

# In Scripts_aSN2N_train.py
config_path = './Config/your_config.json' 

Run the training script:

python Scripts_aSN2N_train.py

The script will:

1. Automatically detect available GPUs.
2. Distribute experiments defined in the JSON config across GPUs.
3. Train the model and perform inference on the test_path data.
4. Save results (images and checkpoints) in the ./images/ directory.

📜 License

This software and corresponding methods can only be used for non-commercial use, and they are under Open Data Commons Open Database License v1.0.