Classification of Multiple H&E Images via an Ensemble Computational Scheme.

Area of Science:

• Computational pathology
• Digital image analysis
• Artificial intelligence in medicine

Background:

• Histopathological image analysis is crucial for disease diagnosis.
• Traditional methods often lack efficiency and accuracy.
• Integrating advanced computational techniques can improve diagnostic capabilities.

Purpose of the Study:

• To develop and validate a computational scheme for classifying histological images.
• To identify optimal combinations of handcrafted fractal descriptors and deep-learned features.
• To enhance computer-aided diagnosis (CADx) systems for histopathology.

Main Methods:

• Utilized multiscale, multidimensional fractal techniques (fractal dimension, lacunarity, percolation) for feature extraction.
• Employed explainable AI (xAI) for quantifying histological images.
• Integrated features from Convolutional Neural Networks (CNNs) including DenseNet-121, EfficientNet-b2, Inception-V3, ResNet-50, and VGG-19.
• Applied a ranking algorithm to identify key feature combinations.
• Validated using heterogeneous ensembles of classifiers (SVM, Naive Bayes, Random Forest, KNN).

Main Results:

• Achieved high classification accuracy rates ranging from 94.83% to 100% across various histological samples (breast cancer, colorectal cancer, oral dysplasia, liver tissue).
• Identified specific pattern ensembles effective for classifying multiple histological image types.
• Demonstrated that optimal performance can be achieved with a reduced feature set (max 25 descriptors).
• Outperformed existing methods reported in the literature.

Conclusions:

• The proposed computational scheme effectively combines fractal descriptors and deep learning for accurate histological image classification.
• This approach offers a robust and efficient method for computer-aided diagnosis.
• The findings provide valuable insights for developing improved diagnostic tools in digital pathology.