Abstract:e12579 Background: Breast cancer is the most common cancer in women worldwide. Although screening has improved early-stage detection, challenges like high false positives and heavy workloads for radiologists persist, emphasizing the importance of early identification of breast irregularities, including masses and micro-calcifications. Methods: We use a subset of the mini-MIAS dataset, randomly choosing 100 abnormal and 100 healthy images, with abnormal images categorized into six sub-types. Preprocessing involves removing noises and excluding background and muscle regions. Our classifier is the ResNet-34 model, a deep convolutional neural network with 34 layers, known for effective pattern identification in medical images. Enhancements include: (1) The Squeeze-and-Excitation (SE) block boosts neural network discriminative power by dynamically recalibrating channel or feature importance. Integrated into ResNet-34, it forms the backbone network, enhancing the identification of relevant patterns for improved diagnostic accuracy. (2) Complex patterns at the system's output are condensed into a common disease pattern through global average pooling (GAP). This reduces model parameters, enhances interpretability, and facilitates the extraction of essential diagnostic information from the entire image. (3) Feature vectors from different layers are concatenated to merge low-level and high-level features. Low-level features encompass fundamental visual attributes such as pixel intensity and texture, while high-level features involve more complex patterns and structures, including organ shapes and anomalies. (4) We employ multiple-way data augmentation techniques to augment the training set, including rotation, translation, scaling, Gamma correction, and shear. The resultant model is referred to as SE-GAP-ResNet. Results: After 10 iterations of 10-fold cross-validation, SE-GAP-ResNet exhibited notable performance with a sensitivity of 95.60%, specificity of 93.30%, precision of 93.49%, accuracy of 94.45%, and an F1 score of 94.51%. Outperforming three state-of-the-art approaches, as shown in the table, two-sample t-tests confirm that SE-GAP-ResNet achieves significantly superior performance (p<0.05). Conclusions: SE-GAP-ResNet excels in identifying breast abnormalities. We aim to enhance the AI system by adding a multi-class diagnosis framework and the capability to predict lesion regions, with the goal of substantially reducing callback rates for abnormal breast diagnoses and improving diagnostic accuracy and efficiency. [Table: see text]

DisepNet for breast abnormality recognition

ResNet-SCDA-50 for Breast Abnormality Classification.

Utilization of DenseNet201 for Diagnosis of Breast Abnormality.

Detection of multiple abnormalities of breast cancer in mammograms using a deep dilated fully convolutional neural network

Abnormal breast identification by nine-layer convolutional neural network with parametric rectified linear unit and rank-based stochastic pooling

SE-GAP-ResNet: A machine learning model for abnormal breast detection.

Abnormality Detection in Mammography using Deep Convolutional Neural Networks

An Automated In-Depth Feature Learning Algorithm for Breast Abnormality Prognosis and Robust Characterization from Mammography Images Using Deep Transfer Learning.

Breast Region Segmentation being Convolutional Neural Network in Dynamic Contrast Enhanced MRI.

Research and Implementation of Breast Cancer Intelligent Recognition Algorithm Based on Deep Convolutional Neural Network

A Hybrid Deep Learning Model for Breast Cancer Detection and Classification

An Introductory Implementation of Breast Cancer Detection from Mammograms and Pixel Intensity with Efficient-Net Other Neural Nets

DEES-breast: deep end-to-end system for an early breast cancer classification

Deep learning-based breast region extraction of mammographic images combining pre-processing methods and semantic segmentation supported by Deeplab v3+

FSE-Net: feature selection and enhancement network for mammogram classification

Dual Convolutional Neural Networks for Breast Mass Segmentation and Diagnosis in Mammography

IDC-Net: Breast cancer classification network based on BI-RADS 4

Histopathological image recognition of breast cancer based on three-channel reconstructed color slice feature fusion

Classification of breast density categories based on SE-Attention neural networks

[Detection of Microcalcification Clusters Regions in Mammograms Combining Discriminative Deep Belief Networks].

Adaptive Channel and Multiscale Spatial Context Network for Breast Mass Segmentation in Full-Field Mammograms