Automating wound assessment: convolutional neural network-based mobile application for SINBAD classification system.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Diabetic foot ulcer (DFU) assessment is crucial for clinical decisions but often lacks specialist access.
• The SINBAD system is a standard for DFU classification.

Purpose of the Study:

• To develop and evaluate a mobile application for automated DFU classification using a lightweight Convolutional Neural Network (CNN).
• To assess the performance of MobileNetV3 Small in classifying five SINBAD components for DFU.

Main Methods:

• A dataset of 996 clinician-labeled DFU images was utilized.
• A MobileNetV3 Small model was trained to classify five SINBAD components.
• Performance was evaluated using accuracy, F1 score, precision, recall, and AUC, and compared to VGG16, ResNet50, and DenseNet121.

Main Results:

• MobileNetV3 Small achieved high F1 scores for Bacterial Infection (93.1%), Area (89.8%), and Neuropathy (86.2%), with excellent recall.
• For Ischemia and Depth, MobileNetV3 Small showed moderate F1 scores (74.4%, 61.6%) and AUCs (84.3%, 80.3%), outperforming VGG16.
• The compact MobileNetV3 Small model demonstrated recall comparable to or exceeding larger models, suitable for sensitive detection.

Conclusions:

• MobileNetV3 Small provides a practical and efficient solution for mobile-based DFU assessment.
• The app's strong recall and compact architecture facilitate deployment in resource-limited settings for consistent SINBAD classification.