MLP-Like Model With Convolution Complex Transformation for Auxiliary Diagnosis Through Medical Images

Area of Science:

• Biomedical image analysis
• Artificial intelligence in healthcare
• Medical diagnostics

Background:

• Facial and tongue images are crucial for intelligence-assisted diagnosis.
• This task is framed as multi-label classification for disease location (DL) and disease nature (DN).
• MLP-like architectures offer computational efficiency and generalization but need enhanced input features.

Purpose of the Study:

• To propose a novel Convolution Complex Transformation MLP-like (CCT-MLP) model.
• To improve multi-label disease location and nature recognition in facial and tongue images.
• To address the limitations of existing MLP-like models in feature extraction.

Main Methods:

• The CCT-MLP model integrates a convolutional Tokenizer and convolutional layers for robust shallow feature extraction.
• It utilizes a Channel-MLP architecture with complex transformations for deep contextual feature extraction.
• Multi-channel features are extracted and fused for multi-label classification.

Main Results:

• The CCT-MLP model demonstrated superior performance on constructed multi-label facial and tongue image datasets.
• Experimental results showed significant improvements in accuracy (Acc) and mean average precision (mAP) compared to existing methods.
• The method effectively compensates for spatial information loss in traditional MLP structures.

Conclusions:

• The CCT-MLP model offers a promising approach for multi-label classification in biomedical image analysis.
• It enhances the diagnostic capabilities of AI systems using facial and tongue imaging.
• The proposed method represents a significant advancement in medical image-based disease recognition.