Histopathological cancer images classification with Deng entropy

Area of Science:

• Digital pathology
• Computational biology
• Medical imaging analysis

Background:

• Histopathological imaging is crucial for tumor detection, diagnosis, and classification.
• Deep learning models like recurrent neural networks (RNNs) and convolutional neural networks (CNNs) have improved digital pathology.
• Existing methods using Tsallis and Shannon entropies face challenges with noise and uncertainty.

Purpose of the Study:

• To classify histopathological cancer images using Deng entropy and bidirectional long short-term memory (bLSTM) networks.
• To develop a novel approach for accurate differentiation between normal and abnormal tissues for pathologists.
• To assess the effectiveness of Deng entropy in capturing complexity and improving cancer classification.

Main Methods:

• Computed Deng entropy using the box covering method at various scales (box sizes).
• Utilized Deng entropy as input vectors for bidirectional LSTM (bLSTM) networks to obtain Deng's information dimensions.
• Analyzed three histopathological datasets: BreakHis (breast), Lung-colon, and PANDA (prostate).

Main Results:

• Achieved high accuracy in binary breast cancer classification (0.98).
• Reached multiclass classification accuracy of 0.99.
• Demonstrated excellent performance for lung (0.98) and colon (0.99) cancer image classification.
• Obtained 0.924 accuracy for prostate cancer image classification.

Conclusions:

• Deng entropy combined with bLSTM networks offers a precise classification system for histopathological images of breast, colon, and lung cancers.
• The proposed methodology effectively mitigates noise and uncertainty, leading to satisfactory cancer classification.
• This innovative approach enhances the ability of pathologists to differentiate between normal and abnormal tissues.