Renal Cancer Detection: Fusing Deep and Texture Features from Histopathology Images.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Histopathological images contain crucial disease markers for diagnosis and prognosis.
• Deep learning (DL) excels with large datasets but struggles with limited medical data.
• Traditional feature extraction offers robustness for smaller datasets, complementing DL.

Purpose of the Study:

• To develop a hybrid classification model for renal cancer detection.
• To integrate deep learning features with traditional texture features for improved accuracy.
• To address the challenge of limited medical datasets in histopathology analysis.

Main Methods:

• A classification model was proposed, fusing features from a deep learning model (Alex-Net) with extracted texture descriptors.
• Five texture feature descriptors from statistic (Histogram of Oriented Gradients, Gray-Level Co-occurrence Matrix, Local Binary Pattern), transform-based (Gabor filters), and model-based (Markov Random Field) families were used.
• The fused features were validated for renal cancer detection using a histopathology dataset.

Main Results:

• The proposed hybrid model demonstrated superior classification performance compared to using Alex-Net alone.
• The fusion approach also outperformed models relying solely on individual texture feature descriptors.
• This indicates the effectiveness of combining deep and texture features for enhanced diagnostic capability.

Conclusions:

• Integrating deep learning features with texture descriptors significantly improves renal cancer detection in histopathology.
• This hybrid approach offers a robust solution for analyzing medical images, especially with limited data.
• The findings highlight the potential of combining machine learning paradigms for advanced computational pathology.