ACTNet: asymmetric convolutional transformer network for diabetic foot ulcers classification.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Diabetic wound diagnostics

Background:

• Existing image classification methods struggle with the complexity and scarcity of data in diabetic foot ulcer (DFU) classification.
• Accurate DFU classification is crucial for effective patient management and treatment.

Purpose of the Study:

• To propose a novel Asymmetric Convolutional Transformer Network (ACTNet) for multi-class (4-class) classification of diabetic foot ulcers.
• To enhance feature extraction and correlation for improved DFU image classification accuracy.

Main Methods:

• Development of an asymmetric convolutional module to model local pixel relationships and extract underlying image features.
• Integration of a novel pooling layer within the Transformer to weight data sequences and correlate features.
• Pre-training the model on ImageNet and fine-tuning on DFU image datasets.

Main Results:

• The ACTNet model achieved an F1-score of 0.593 and an AUC value of 0.824 on the DFUC2021 test set.
• Demonstrated excellent performance in classifying diabetic foot ulcers despite data scarcity.
• The asymmetric convolutional module effectively guided the network to focus on informative image regions.

Conclusions:

• ACTNet presents a promising approach for accurate diabetic foot ulcer classification, even with limited datasets.
• The proposed architectural components enhance the network's expressive and correlative capabilities.
• This work contributes to advancing AI-driven diagnostic tools for diabetic complications.