NMC Code

📚 Related Paper

This code is an implementation for the following paper:

A Comparative Study of Lesion-Centered and Severity-Based Approaches to Diabetic Retinopathy Classification: Improving Interpretability and Performance

• Journal: Biomedicines (MDPI)
• DOI: https://doi.org/10.3390/biomedicines13061446
• Full Paper: https://www.mdpi.com/2227-9059/13/6/1446

---

Overview

This repository contains the refactored code for the NMC system. The code has been cleaned by removing unnecessary outputs and duplicate code, keeping only the core functionality.

📁 Directory Structure

nmc_clean/
├── core/           # Core NMC modules
├── configs/        # Configuration files
├── notebooks/      # Refactored Jupyter notebooks
├── tools/          # Evaluation tools used in NMCS
└── README.md       # This file

🔧 Core Modules

Models

• EfficientNetV2: EfficientNetV2-based models
  • EfficientNetV2MModel: Single-label classification
  • EfficientNetV2MModelMulti: Multi-label classification
• ResNet: ResNet-based models
  • ResNet50Model: Single-label classification
  • ResNet50MultiHeadModel: Multi-label classification
• FGMaxxVit: FGMaxxVit-based models
  • FGMaxxVit: Single-label classification
  • FGMaxxVit_Multi: Multi-label classification
• TestCNN: Simple CNN model for testing

Utils

• augmentations.py: Data augmentation functions
• losses.py: Loss functions
• metrics.py: Evaluation metrics
• optimizers.py: Optimization algorithms
• schedulers.py: Learning rate schedulers
• utils/: Other utility functions

📊 Configuration Files

• NMC.yaml: NMC dataset training configuration
• APTOS.yaml: APTOS dataset training configuration
• ODIR.yaml: ODIR dataset training configuration
• Multi_Task.yaml: Multi-task learning configuration

📓 Jupyter Notebooks (Core NMCS Files)

NMC Related

• NMC.ipynb: Basic NMC model training and evaluation
• NMC_singlelabel.ipynb: Single-label NMC training
• NMC_labelchain.ipynb: Label chain-based NMC training
• NMC_confusion.ipynb: Confusion matrix analysis

APTOS Related

• APTOS.ipynb: Basic APTOS model training
• APTOS_singlelabel.ipynb: Single-label APTOS training
• APTOS_NMC_finetuning.ipynb: APTOS fine-tuning with NMC
• NMC_APTOS_finetuning.ipynb: NMC fine-tuning with APTOS

Visualization and Analysis

• NMC_APTOS_visualization.ipynb: Visualization tools
• NMC_APTOS_gradcam.ipynb: Grad-CAM analysis
• NMC_APTOS_OSM.ipynb: OSM (Object Saliency Map) analysis

Special Models

• NMC_APTOS_BIFPN.ipynb: BIFPN (Bidirectional Feature Pyramid Network) model
• NMC_APTOS_FPN.ipynb: FPN (Feature Pyramid Network) model

🛠️ Tools (Actually Used in NMCS)

• val.py: Model evaluation tool (includes evaluate_epi function)
• episodic_utils.py: Episodic learning utilities

🚀 Quick Start

1. Environment Setup

   pip install -r requirements.txt

   Or install individually:

   pip install torch torchvision torchaudio
   pip install numpy pandas scipy scikit-learn
   pip install opencv-python Pillow matplotlib seaborn
   pip install tqdm PyYAML tabulate jupyter

2. Configuration Check

   • Check data paths and model settings in configs/NMC.yaml

3. Run Notebooks

   • Execute desired notebooks from the notebooks/ folder

⚠️ Important Notes

• This code is a refactored version that includes only the core functionality actually used in NMCS
• Original code remains unchanged
• Unnecessary files have been removed (contrastive_proto, multi_task, etc.)
• Set correct paths in configuration files before execution
• GPU environment is required (CUDA support)

🔗 Dependencies

• PyTorch
• torchvision
• scikit-learn
• matplotlib
• seaborn
• pandas
• numpy
• PIL (Pillow)
• OpenCV
• tabulate
• tqdm
• PyYAML

📝 License

Follows the license of the original project.

---

This code is a refactored version containing only the core functionality of the NMC system.