Breast cancer histopathological image classification using convolutional neural networks with small SE-ResNet module

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Accurate breast cancer diagnosis from histopathological images is crucial but relies heavily on radiologist expertise, leading to potential discrepancies.
• Computer-aided diagnosis (CAD) systems offer a valuable tool to enhance the reliability and consistency of expert decision-making in pathology.

Purpose of the Study:

• To develop and evaluate a novel convolutional neural network (CNN) for the automatic classification of breast cancer histology images.
• To improve the precision and efficiency of identifying malignant tumors and their subtypes in histopathological images.
• To enhance diagnostic reliability in breast cancer assessment through advanced AI techniques.

Main Methods:

• Design of a novel CNN incorporating a convolutional layer, a small SE-ResNet module (combining residual module and Squeeze-and-Excitation block), and a fully connected layer.
• Development of a new learning rate scheduler to optimize model performance without complex fine-tuning.
• Application of the model to the BreakHis dataset for binary (benign/malignant) and multi-class (eight subtypes) classification of breast cancer histology images.

Main Results:

• The proposed CNN model achieved high accuracy in binary classification, ranging from 98.87% to 99.34%.
• For multi-class classification into eight subtypes, the model demonstrated strong performance with accuracies between 90.66% and 93.81%.
• The small SE-ResNet module provided comparable performance to larger models with significantly fewer parameters.

Conclusions:

• The developed CNN, featuring a novel SE-ResNet module and learning rate scheduler, offers an effective solution for automated breast cancer histology image classification.
• The model's high accuracy in both binary and multi-class settings demonstrates its potential to assist pathologists and improve diagnostic outcomes.
• This AI-driven approach holds promise for enhancing the consistency and precision of breast cancer diagnosis in clinical practice.