Medical Image Classification Based on Deep Features Extracted by Deep Model and Statistic Feature Fusion with...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence

Background:

• Traditional medical image classification methods struggle with high-level concept representation and generalization.
• Deep learning models face challenges with high-resolution medical images and limited datasets, leading to high computational costs.

Purpose of the Study:

• To develop an efficient deep learning model for medical image classification that overcomes limitations of existing methods.
• To improve the accuracy and generalization ability of computer-aided diagnosis systems.

Main Methods:

• A deep convolutional neural network was trained as a coding network to extract high-level feature vectors from raw medical image pixels.
• Selected traditional features were extracted based on medical image domain knowledge.
• A neural network-based model (CNMP) was designed to fuse deep learning and traditional features.

Main Results:

• The proposed CNMP model achieved 90.1% accuracy on the HIS2828 dataset.
• The model achieved 90.2% accuracy on the ISIC2017 dataset.
• These results surpass current successful methods in medical image classification.

Conclusions:

• The CNMP model effectively combines deep learning and traditional features for superior medical image classification.
• This approach enhances the performance of computer-aided diagnosis systems, offering a promising direction for future research.