Explainable Transfer Learning Ensemble with Dynamic Weighting for Kidney Disease Classification in CT Images

Deep learning ensemble framework for automated kidney disease classification from CT scans with explainable AI visualizations using GradCAM and LIME.

Overview

Kidney-XAI implements an ensemble of deep learning models (MobileNetV2, ResNet50, EfficientNet-B1) for multi-class kidney disease classification. The framework provides interpretable predictions through gradient-based and perturbation-based explainability techniques.

Architecture

Figure: Ensemble architecture combining MobileNetV2, ResNet50, and EfficientNet-B1 with dynamic weighting and explainability modules.

Dataset

Dataset Structure

data/
├── Cyst/
│   ├── image_001.jpg
│   └── ...
├── Normal/
│   ├── image_001.jpg
│   └── ...
├── Stone/
│   ├── image_001.jpg
│   └── ...
└── Tumor/
    ├── image_001.jpg
    └── ...

Installation

Requirements

• Python 3.8+
• PyTorch 2.0+
• CUDA 11.0+ (for GPU training)

Setup

git clone https://github.com/darrinbright/Kidney-XAI.git
cd Kidney-XAI

conda create -n kidney-xai python=3.8
conda activate kidney-xai

pip install -r requirements.txt

Usage

Training

python train.py --data_dir ./data --epochs 25 --batch_size 32

Training Options

python train.py \
  --data_dir ./data \
  --epochs 25 \
  --batch_size 32 \
  --lr_mobilenet 8e-5 \
  --lr_resnet 4e-5 \
  --lr_efficientnet 1.5e-4 \
  --weight_decay 5e-4 \
  --focal_gamma 2.0 \
  --label_smoothing 0.1 \
  --patience 5 \
  --save_dir ./checkpoints

Key Hyperparameters

Parameter	Default	Description
--epochs	25	Number of training epochs
--batch_size	32	Batch size for training
--focal_gamma	2.0	Focal loss gamma parameter
--label_smoothing	0.1	Label smoothing factor
--patience	5	Early stopping patience

Explainability

GradCAM Visualization

Gradient-weighted Class Activation Mapping highlights discriminative regions:

# GradCAM is automatically generated during training
# Visualizations saved to ./results/GradCAM/

LIME Explanation

Local Interpretable Model-agnostic Explanations for prediction understanding:

# LIME visualizations saved to ./results/LIME/

Project Structure

Kidney-XAI/
├── train.py                 # Main training script
├── requirements.txt         # Python dependencies
├── README.md               # Documentation
├── .gitignore              # Git ignore rules
├── data/                   # Dataset directory      
└── results/                # Outputs and visualizations
    ├── GradCAM/
    ├── LIME/
    ├── ConfusionMatrix/
    ├── ROC/
    └── TrainingValidation/

Citation

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

@software{kidney_xai_2026,
  title={Explainable Transfer Learning Ensemble with Dynamic Weighting for Kidney Disease Classification in CT Images},
  author={Darrin Bright, Praveen Kumar Reddy},
  year={2026},
  url={https://github.com/darrinbright/Kidney-XAI}
}

Contact

For questions or collaborations, please contact: darrin.bright2022@vitstudent.ac.in