Double Attention: An Optimization Method for the Self-Attention Mechanism Based on Human Attention.

Area of Science:

• Biomedical engineering
• Artificial intelligence
• Medical diagnostics

Background:

• Artificial intelligence (AI) is increasingly integrated into daily life.
• The self-attention mechanism and Transformer architecture are pivotal in AI advancements.
• AI shows promise as a precise diagnostic and predictive tool in medicine.

Purpose of the Study:

• To enhance the accuracy and biomimetic performance of neural networks in medical applications.
• To introduce the Double-Attention (DA) method for improved information acquisition in AI models.
• To validate the effectiveness of DA on benchmark datasets and real-world patient data.

Main Methods:

• Proposed the Double-Attention (DA) method, enhancing the self-attention mechanism.
• Incorporated matrices from shifted images to enable preemptive information acquisition.
• Applied the DA method to hospital-collected kidney datasets for diabetes diagnosis.

Main Results:

• Demonstrated superior performance of the DA method across various benchmark datasets.
• Achieved high accuracy in diabetes diagnosis using patient kidney datasets.
• Significantly reduced computational demands compared to existing methods.

Conclusions:

• The DA method significantly improves neural network performance by enhancing biomimetic attention.
• DA shows strong potential for developing innovative, bioinspired diagnostic tools in healthcare.
• The approach offers an effective and efficient solution for medical diagnosis, particularly for kidney-related conditions.