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

Breast cancer is a major health threat, with early detection crucial for improving cure and survival rates. Current systems rely on imaging technology, but digital pathology and computerized analysis can enhance accuracy, reduce false predictions, and improve medical care for breast cancer patients. The study explores the challenges in identifying benign and malignant breast cancer lesions using microscopic image datasets. It introduces a low-dimensional multiple-channel feature-based method for breast cancer microscopic image recognition, overcoming limitations in feature utilization and computational complexity. The method uses RGB channels for image processing and extracts features using level co-occurrence matrix, wavelet, Gabor, and histogram of oriented gradient. This approach aims to improve diagnostic efficiency and accuracy in breast cancer treatment. The core of our method is the SqE-DDConvNet algorithm, which utilizes a 3 x 1 convolution kernel, SqE-DenseNet module, bilinear interpolation, and global average pooling to enhance recognition accuracy and training efficiency. Additionally, we incorporate transfer learning with pre-trained models, including mVVGNet16, EfficientNetV2B3, ResNet101V2, and CN2XNet, preserving spatial information and achieving higher accuracy under varying magnification conditions. The method achieves higher accuracy compared to baseline models, including texture and deep semantic features. This deep learning-based methodology contributes to more accurate image classification and unique image recognition in breast cancer microscopic images.Research Highlights Introduces a low-dimensional multiple-channel feature-based method for breast cancer microscopic image recognition. Uses RGB channels for image processing and extracts features using level co-occurrence matrix, wavelet, Gabor, and histogram of oriented gradient. Employs the SqE-DDConvNet algorithm for enhanced recognition accuracy and training efficiency. Transfer learning with pre-trained models preserves spatial information and achieves higher accuracy under varying magnification conditions. Evaluates predictive efficacy of transfer learning paradigms within microscopic analysis. Utilizes CNN-based pre-trained algorithms to enhance network performance. Deep convolutional neural network (CNN)-based architectures involves pretraining models, preprocessing data, enhancing & augmenting. image