Gated Backbone Fusion with Transformer Encoder for Diabetic Foot Osteomyelitis Screening and Localization in...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Diabetic Complications

Background:

• Diabetic foot osteomyelitis (DFO) is a major cause of lower-limb complications in diabetes patients.
• Early detection is crucial to prevent severe outcomes like amputation.
• Conventional radiography often misses subtle DFO signs, and current AI models struggle with localized patterns.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework, DualBack-GFT, for automated detection and localization of DFO in plain radiographs.
• To improve the accuracy and robustness of DFO assessment compared to existing methods.

Main Methods:

• Proposed DualBack-GFT framework using EfficientNet-B6 and ResNet-50 backbones with gated fusion.
• Incorporated transformer encoders to model long-range dependencies.
• Implemented a two-stage approach: binary classification and confidence-weighted bounding-box localization.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.9683 for DFO detection.
• Demonstrated an average ground truth coverage of 62.71% for localization.
• Outperformed established baseline models on a curated expert-annotated dataset.

Conclusions:

• DualBack-GFT shows significant potential for accurate and interpretable DFO assessment in clinical settings.
• The dual-stage, attention-enhanced architecture effectively captures localized pathological patterns.
• This AI approach can aid clinicians in timely DFO diagnosis and management, potentially reducing amputations.