# A Machine Learning Model for Early Detection of Diabetic Foot using Thermogram Images

**Authors:** Amith Khandakar, Muhammad E. H. Chowdhury, Mamun Bin Ibne Reaz, Sawal Hamid Md Ali, Md Anwarul Hasan, Serkan Kiranyaz, Tawsifur Rahman, Rashad Alfkey, Ahmad Ashrif A. Bakar, Rayaz A. Malik  
**arXiv:** [2106.14207](https://arxiv.org/abs/2106.14207)  
**Date:** 2021-06-27  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2106.14207v1/

## Key points

- Diabetes foot ulceration (DFU) and amputation are significant contributors to morbidity among diabetic patients.
- Early detection and prevention of DFU are crucial in reducing the risk of complications.
- Thermogram images show potential in detecting changes in plantar temperature, serving as an early indicator of DFU development.
- Challenges exist due to the heterogeneous distribution of plantar temperature, making accurate quantification difficult for predictive purposes.
- A team of researchers led by Amith Khandakar and Muhammad E. H. Chowdhury developed a machine learning model for early detection of diabetic foot using thermogram images.
- The study compared various Convolutional Neural Networks (CNNs) and found that MobilenetV2 achieved an F1 score of approximately 95% while AdaBoost Classifier achieved an even higher F1 score of 97%.
- The algorithm developed by the researchers can be deployed as a smartphone application, allowing users to monitor DFU progression at home for timely interventions.
- This research represents a significant advancement in leveraging machine learning techniques for early detection and monitoring of diabetic foot conditions using thermogram images, potentially improving outcomes for diabetic patients at risk of DFU and amputation.

## Summary

Diabetes foot ulceration (DFU) and amputation are significant contributors to morbidity among diabetic patients. The early detection and prevention of DFU are crucial in reducing the risk of complications. Recent studies have highlighted the potential of thermogram images in detecting changes in plantar temperature, which can serve as an early indicator of DFU development. However, the heterogeneous distribution of plantar temperature poses challenges in accurately quantifying and utilizing this information for predictive purposes. In response to this challenge, a team of researchers led by Amith Khandakar, Muhammad E. H. Chowdhury, and their colleagues have developed a machine learning model aimed at early detection of diabetic foot using thermogram images. By leveraging a combination of machine learning-based scoring techniques, feature selection, optimization methods, and learning classifiers, the researchers compared various state-of-the-art Convolutional Neural Networks (CNNs) to identify an effective solution for diabetic foot identification. Their study found that a comparatively shallow CNN model known as MobilenetV2 achieved an impressive F1 score of approximately 95% for classifying two-feet thermogram images. Additionally, the AdaBoost Classifier utilized 10 features and achieved an even higher F1 score of 97%. These results demonstrate the efficacy of their proposed algorithm in accurately identifying diabetic foot conditions based on thermogram images. Furthermore, the researchers conducted a comparison of inference times for the best-performing networks and confirmed that their algorithm could be deployed as a smartphone application. This innovative approach allows users to monitor the progression of DFU in a home setting, providing valuable insights into potential risks and enabling timely interventions to prevent complications. Overall, this research represents a significant advancement in leveraging machine learning techniques for early detection and monitoring of diabetic foot conditions using thermogram images. By offering a robust solution with high accuracy rates and practical applicability through smartphone technology, this study has the potential to improve outcomes for diabetic patients at risk of DFU and amputation.