Advancing diabetic foot ulcer care with artificial intelligence

This breakthrough could give researchers a better understanding of diabetic foot ulcers and spur the development of new treatments.

TORONTO–A new study from the KITE Research Institute could advance diabetic foot ulcer (DFU) treatment and diagnosis.  

DFU is an open sore or wound located on the foot of a person living with diabetes. This condition can lead to amputation and in some cases death. 

The team at KITE created a new method of gathering DFU data and trained a deep machine learning program to detect the condition. This breakthrough could give researchers a better understanding of DFU and spur the development of new treatments. 

The results of their research were published in a special ICAIR edition of Biomedical Engineering OnLine.

KITE trainee and the paper’s first author Reza Basiri discussed his team’s findings with the ICAIR organizing committee.

Which patient groups are most affected by this? 

Patients with diabetes foot complications, mainly a condition called diabetic foot ulcer (DFU), are mostly affected. This condition predominantly occurs in adults over the age of 65 who have vascular insufficiency coupled with diabetes. A DFU is a deep open wound that requires early and multidisciplinary care, where acute conditions can lead to amputations or even death.

 What did you find? 

Employing this protocol, we successfully curated the extensive Zivot DFU dataset over a year, comprising about 3,700 images from 269 participants and presented a summary of clinical and demographical features for this patient population. Additionally, high-performance metrics were achieved by training the selected EfficientNetb3-UNet deep learning (DL) network on this dataset, with an F1-score of 0.79 and mAP of 0.86.

 Why does this matter? 

Research and development for DFU diagnosis and treatment have been limited for two reasons: 1. Absence of large holistic DFU datasets, and 2. The incomparability of available datasets is due to various data collection methods. To this end, we created a comprehensive and large DFU dataset using our proposed unifying multimodal protocol, called Zivot protocol and dataset.

 What is the potential impact? 

The study established a standardized framework for DFU data collection, and a baseline for DFU detection using a UNet architecture on the Zivot dataset. These outcomes underscore the significance of the protocol and dataset in advancing DL-based DFU research and development.

Research Spotlight: 
Protocol for metadata and image collection

KITE Research Institute trainee, Ph.D. candidate at the Institute of Biomedical Engineering at the University of Toronto

Name of Publication:
Protocol for metadata and image collection at diabetic foot ulcer clinics: enabling research in wound analytics and deep learning

Name of Journal:
BioMedical Engineering OnLine