Ensemble of Handcrafted and Learned Features for Colorectal Cancer Classification.

Area of Science:

• Medical image analysis
• Computational pathology
• Artificial intelligence in oncology

Background:

• Colorectal cancer (CRC) diagnosis relies on histopathology, which is time-consuming and subjective.
• Current deep learning methods for CRC classification require extensive annotated data and lack interpretability.
• Handcrafted texture descriptors offer domain insights but may miss complex patterns.

Purpose of the Study:

• To develop a robust and interpretable automated classification system for colorectal cancer.
• To integrate the strengths of handcrafted texture descriptors and deep learning features.
• To enhance the accuracy and reliability of CRC diagnosis using a hybrid approach.

Main Methods:

• An ensemble model was developed integrating handcrafted color texture features with deep learning features from Convolutional Neural Networks (CNNs).
• The approach aimed to create a more discriminative and robust feature space by combining complementary information.
• Performance was evaluated using standard metrics against state-of-the-art methods.

Main Results:

• The proposed ensemble method achieved a high accuracy of 99.20% in CRC classification.
• The integration of color textures and deep learning features significantly improved classification performance.
• The ensemble approach demonstrated superior results compared to existing state-of-the-art techniques.

Conclusions:

• Integrating handcrafted texture descriptors with deep learning features offers a powerful strategy for medical image analysis.
• This hybrid approach significantly advances automated colorectal cancer classification.
• The study highlights the potential for improved diagnostic accuracy and efficiency in computational pathology.