CNN Attention Guidance for Improved Orthopedics Radiographic Fracture Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Convolutional neural networks (CNNs) are increasingly used for fracture classification in orthopedic imaging.
• A major limitation of CNNs is their lack of transparency, hindering clinical trust and adoption, especially with limited medical data.
• While visualizing CNN attention is explored, using it to enhance network learning is less investigated.

Purpose of the Study:

• To investigate the effectiveness of incorporating human-provided attention guidance into CNNs for orthopedic fracture classification.
• To determine if explicit human guidance can direct network attention and improve diagnostic performance.

Main Methods:

• The study proposes a method to regularize CNNs using human attention guidance, directing the network to focus on specific image regions.
• Experiments were conducted on two orthopedic radiographic fracture classification datasets.

Main Results:

• Explicit human-guided attention effectively directed the CNN's focus to relevant areas in the radiographic images.
• This attention guidance significantly improved the classification performance of the CNNs.

Conclusions:

• Human attention guidance is a viable strategy to enhance the learning process and diagnostic accuracy of CNNs in orthopedic imaging.
• This approach addresses the transparency issue and can boost the clinical utility of AI in fracture detection.