Breast Cancer Pathological Image Classification Based on the Multiscale CNN Squeeze Model.

Area of Science:

• * Digital Pathology
• * Medical Imaging Analysis
• * Artificial Intelligence in Oncology

Background:

• * Automated histopathological image identification is crucial for efficient and accurate cancer diagnosis.
• * Computerized breast cancer multiclassification using histological images remains an under-explored area.
• * Existing methods require improvement in accuracy and efficiency for clinical application.

Purpose of the Study:

• * To develop a deep learning-based strategy for automated breast cancer pathology image classification.
• * To introduce a multiscale channel recalibration model (msSE-ResNet) to enhance classification accuracy.
• * To evaluate the performance of the proposed model on a public dataset for breast cancer classification.

Main Methods:

• * Development of a channel refinement model incorporating an attention mechanism to reduce superfluous features.
• * Implementation of a multiscale channel recalibration model to improve feature recalibration accuracy.
• * Construction of the msSE-ResNet convolutional neural network, integrating multiscale properties and channel recalibration.

Main Results:

• * The msSE-ResNet model achieved a classification accuracy of 88.87% for benign/malignant breast pathology images across various magnifications.
• * The model demonstrated resilience with diseased images.
• * Comparative analysis indicated that spatial recalibration models performed poorly on this specific task.

Conclusions:

• * The msSE-ResNet deep learning model shows significant promise for automated breast cancer classification from histopathological images.
• * The proposed multiscale channel recalibration approach effectively enhances classification accuracy.
• * This automated system can potentially expedite diagnoses and reduce errors in breast cancer pathology.