Parallel Structure Deep Neural Network Using CNN and RNN with an Attention Mechanism for Breast Cancer Histology...

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Accurate classification of hematoxylin-eosin-stained breast biopsy images is crucial for effective cancer diagnosis and treatment.
• Existing deep learning models often employ serial architectures for feature extraction, which may not capture complex image patterns optimally.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for classifying breast biopsy images into four distinct categories: normal tissues, benign lesions, in situ carcinomas, and invasive carcinomas.
• To enhance image feature extraction and classification accuracy using a parallel CNN-RNN structure and advanced deep learning techniques.

Main Methods:

• A parallel deep learning architecture combining a convolutional neural network (CNN) and a recurrent neural network (RNN) for simultaneous image feature extraction.
• Integration of a perceptron attention mechanism, adapted from natural language processing, to unify features from CNN and RNN components.
• Implementation of switchable normalization in convolution layers and targeted dropout in fully connected layers for improved model performance and regularization.
• Utilized model fusion and test-time augmentation on three distinct breast biopsy image datasets during the testing phase.

Main Results:

• The proposed deep learning model demonstrated superior performance in classifying breast biopsy images compared to existing state-of-the-art methods.
• The parallel CNN-RNN structure, coupled with the attention mechanism, effectively captured complex image features for accurate classification.
• Advanced techniques like switchable normalization, targeted dropout, model fusion, and test-time augmentation contributed to the model's enhanced accuracy.

Conclusions:

• The novel deep learning model offers a significant advancement in the automated classification of breast biopsy images.
• The parallel architecture and specialized mechanisms provide a robust framework for improving diagnostic accuracy in computational pathology.
• This approach holds potential for enhancing the efficiency and reliability of breast cancer diagnosis.