Related Experiment Video
Updated: Jan 7, 2026
06:27
Automated Analysis of C. elegans Fluorescence Images using SegElegans
Published on: October 10, 2025
530
Automated DFU detection through GA-selected CNN ensemble with Grad-CAM interpretability
Shreya Girotra1, Achin Jain1, Sarita Yadav1
1Department of Information Technology, Bharati Vidyapeeth's College of Engineering, New Delhi, India.
Scientific Reports
|December 14, 2025
Summary
This study introduces an optimized deep learning model for detecting diabetic foot ulcers (DFU). The model achieved 97% accuracy, improving detection and interpretability for better patient outcomes.
Area of Science:
- Medical image analysis
- Artificial intelligence in healthcare
- Diabetic complications
Background:
- Diabetic foot ulcers (DFU) are a severe complication of diabetes, often leading to amputation and significant mortality.
- Current deep learning models for DFU detection face challenges in optimizing accuracy and interpretability.
- Advancements in machine learning offer potential for improved DFU detection through medical image analysis.
Purpose of the Study:
- To develop and validate an optimized deep learning model for accurate and interpretable detection of diabetic foot ulcers (DFU).
- To enhance the performance of DFU detection models by integrating Convolutional Neural Network (CNN) structures with Genetic Algorithm (GA) ensembles.
- To improve the transparency of DFU detection models using visualization techniques like Grad-CAM.
Main Methods:
- Developed a custom Convolutional Neural Network (CNN) architecture trained with seven different optimizers.
- Utilized a Genetic Algorithm (GA) to create an ensemble model from the best-performing individual CNN models.
- Implemented Grad-CAM (Gradient-weighted Class Activation Mapping) for model interpretability and decision transparency.
Main Results:
- The GA-based ensemble model achieved high performance metrics: 97% accuracy, 95% precision, 99% recall, and 97% F1 score.
- The ensemble model outperformed individual single models in DFU detection accuracy and robustness.
- Grad-CAM successfully visualized ulcer-associated features, providing insights into the model's decision-making process.
Conclusions:
- The proposed optimized deep learning model, combining CNNs and GA ensembles, significantly improves DFU detection accuracy and interpretability.
- Grad-CAM enhances model transparency, offering valuable insights for healthcare professionals in clinical decision-making.
- This approach represents a promising advancement in leveraging AI for managing diabetic foot complications.
