Advanced feature learning and classification of microscopic breast abnormalities using a robust deep transfer...

Area of Science:

• Digital pathology
• Medical imaging analysis
• Computational biology

Background:

• Early breast cancer detection is vital for survival.
• Current imaging methods have limitations.
• Digital pathology and AI can improve accuracy.

Purpose of the Study:

• To develop an accurate method for classifying benign and malignant breast cancer lesions from microscopic images.
• To address challenges in feature extraction and computational complexity.

Main Methods:

• A low-dimensional, multiple-channel feature-based approach using RGB channels.
• Feature extraction via co-occurrence matrix, wavelet, Gabor, and histogram of oriented gradients.
• The SqE-DDConvNet algorithm with transfer learning (mVVGNet16, EfficientNetV2B3, ResNet101V2, CN2XNet).

Main Results:

• The proposed method achieved higher accuracy than baseline models.
• Transfer learning preserved spatial information and improved accuracy across magnifications.
• Enhanced recognition accuracy and training efficiency were observed.

Conclusions:

• The deep learning methodology offers more accurate image classification for breast cancer microscopic images.
• This approach contributes to improved diagnostic efficiency and patient care.
• The study validates the efficacy of transfer learning in microscopic image analysis.