A Novel Multispace Image Reconstruction Method for Pathological Image Classification Based on Structural Information.

Area of Science:

• Biomedical imaging
• Computational pathology
• Machine learning in medicine

Background:

• Pathological image classification is crucial for accurate diagnosis and grading in various medical applications.
• Existing methods often face challenges in effectively extracting comprehensive features from complex pathological images.

Purpose of the Study:

• To develop a novel classification framework for pathological images.
• To enhance feature extraction and improve classification accuracy using multispace image reconstruction and transfer learning.

Main Methods:

• A multispace image reconstruction technique was employed to create a three-channel image (gradient, GLCM, LBP).
• Pretrained VGG-16 network was used for feature extraction from both original (RGB) and reconstructed images.
• Long short-term memory (LSTM) layer was utilized for feature selection and refinement, followed by softmax classification.

Main Results:

• The proposed framework demonstrated superior performance compared to existing methods on a malignant lymphoma dataset.
• The integration of multispace reconstruction significantly improved the discriminative capability of extracted features.
• Experimental results validated the effectiveness of the novel classification approach.

Conclusions:

• The developed framework offers a promising approach for accurate pathological image classification.
• Multispace image reconstruction combined with transfer learning and LSTM enhances feature representation for biomedical applications.
• This method holds potential for improving diagnostic accuracy in clinical settings.