Related Experiment Video
Updated: Jan 7, 2026
Fabrication and Characterization of a Conformal Skin-like Electronic System for Quantitative, Cutaneous Wound Management
Published on: September 2, 2015
Automating wound assessment: convolutional neural network-based mobile application for SINBAD classification system
Farideh Mostafavi1, Sujit Kumar Das2, Mohammad Reza Amini3
1Department of Epidemiology, School of Public Health and Safety, Shahid Beheshti University of Medical Sciences, Shahid Shahriari Sq., Student Blvd., Valenjak, Tehran, 1983969411 Iran.
Introduction:
Diabetic foot ulcer (DFU) assessment using the SINBAD system is essential for clinical decision-making but often limited by access to specialists. This study presents a mobile application powered by a lightweight Convolutional Neural Networks (MobileNetV3 Small) to automate DFU classification.
Methods:
A dataset of 996 clinician-labeled DFU images was used to train the model to classify five SINBAD components. Model performance was evaluated using accuracy, F1 score, precision, recall, and AUC, and compared against VGG16, ResNet50, and DenseNet121.
Results:
MobileNetV3 demonstrated strong performance across most SINBAD components, achieving high F1 scores for Bacterial Infection (93.1%), Area (89.8%), and Neuropathy (86.2%), along with excellent recall (96.2%, 98.2%, and 94.5%, respectively). In Ischemia and Depth classification, MobileNetV3 achieved moderate F1 scores (74.4% and 61.6%) and AUCs (84.3% and 80.3%), outperforming VGG16 and closely approaching the performance of larger models like DenseNet121. Notably, despite its compact size, MobileNetV3 often matched or exceeded the recall of more complex models, indicating its suitability for sensitive clinical detection tasks.
Conclusion:
MobileNetV3 offers a practical, efficient solution for mobile-based DFU assessment. Its strong recall and compact architecture support deployment in outpatient or resource-limited settings for consistent SINBAD classification.