Performance Evaluation of Deep Learning Models for Image Classification Over Small Datasets: Diabetic Foot Case Study

This project is designed to replicate the experiments of the paper "Performance Evaluation of Deep Learning Models for Image Classification Over Small Datasets: Diabetic Foot Case Study". Because the data used in the work is not public, this repository uses MedMNIST Breast Ultrasound as dataset. In order to reproduce the data scarcity problem, a random undersampling is applied to the dataset.

Prerequisites

• Anaconda
• Git
• PyTorch
  • version 1.8.0 or above
• Torchvision
• Matplotlib
• Pandas

Submodules

This repository contains the following submodules:

• IPDL
• MedMNIST

In order to download the submodules in the cloning process, use the following instruction:

git clone --recurse-submodules [repository url]

References

If you find our library useful in your research, please consider citing us:

@article{hernandez2022performance,
  title={Performance Evaluation of Deep Learning Models for Image Classification Over Small Datasets: Diabetic Foot Case Study},
  author={Hernandez-Guedes, Abian and Santana-Perez, Idafen and Arteaga-Marrero, Natalia and Fabelo, Himar and Callico, Gustavo M and Ruiz-Alzola, Juan},
  journal={IEEE Access},
  volume={10},
  pages={124373--124386},
  year={2022},
  publisher={IEEE}
}

TODO

• Explaining that MedMNIST Breast Ultrasound is used
• Explanation of the notebooks