Multi-Class Classification of Breast Cancer Subtypes Using ResNet Architectures on Histopathological Images.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Breast cancer diagnosis relies on histopathology, which is subjective and time-consuming.
• Accurate classification of breast cancer subtypes is crucial for effective treatment and patient outcomes.
• Existing deep learning models often focus on binary classification, limiting subtype differentiation.

Purpose of the Study:

• To develop and evaluate a deep learning framework for multi-class classification of eight breast cancer histopathological subtypes.
• To compare the performance of different ResNet architectures (ResNet-18, ResNet-34, ResNet-50) for this task.
• To assess the effectiveness of transfer learning and data augmentation in improving classification accuracy.

Main Methods:

• Utilized convolutional neural networks (CNNs), specifically ResNet architectures (ResNet-18, ResNet-34, ResNet-50), pre-trained on ImageNet.
• Implemented extensive data augmentation techniques to enhance model robustness across different magnifications.
• Performed multi-class classification to distinguish between four benign and four malignant breast cancer subtypes.

Main Results:

• ResNet-50 demonstrated superior performance, achieving a maximum accuracy of 92.42%.
• The model attained an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 99.86% and an average specificity of 98.61%.
• The framework successfully captured fine-grained histopathological features for accurate subtype classification.

Conclusions:

• Deep learning, particularly CNNs with transfer learning, offers a powerful tool for accurate multi-class breast cancer subtype classification.
• The proposed framework significantly improves upon traditional methods by reducing subjectivity and enhancing diagnostic efficiency.
• These findings support the clinical utility of AI-driven histopathological analysis for breast cancer diagnosis.