Early breast cancer detection, affected cell classification, and segmentation framework

Introduction

Worldwide, breast cancer (BC) is the leading cause of cancer-related deaths among women. Early detection significantly enhances the likelihood of successful treatment and long-term survival.

Objective

This study proposes a novel breast cancer detection framework based on transfer learning (TL) techniques to assist in efficient automated diagnosis, stage classification, and segmentation of the affected cells.

Methods

The framework utilizes mammography and ultrasound images to increase its applicability in diverse clinical settings. The collected datasets are preprocessed and prepared using the Medical Image Processing, Analysis, and Visualization (MIPAV) software with the bilateral filtering technique. Classification of breast cancer into three stages—normal, benign, and malignant—is achieved through a multi-class classifier, EfficientVGG-Net v1, which integrates the EfficientNetV2B0 and Visual Geometry Group (VGG16) TL models. Moreover, a binary segmentation model, termed B²CS-AResu-Net (Binary Breast Cancer Cell Segmentation using Attention Residual U-Net), is proposed to segment regions affected by cancerous cells.

Results

Empirical evaluations demonstrate that the proposed framework classification and segmentation models yield highly competitive performance across multiple evaluation metrics. The EfficientVGG-Net v1 model achieves classification results of 99.27 % for accuracy, precision, and recall, 99.87 % for f1-score, and 99.24 % for the area under the curve. Besides, the B²CS-AResu-Net segmentation model attains 99.45 %, 99.42 %, 99.34 %, 99.44 %, and 99.42 % for the aforementioned classification metrics. Segmentation quality is further validated by achieving a 95.98 % Dice Similarity Coefficient, 92.29 % Intersection over Union, and 97.45 % Matthews Correlation Coefficient.

Conclusion

The proposed framework substantially improves the dependability of artificial intelligence-assisted diagnostic tools for breast cancer, enhancing clinical decision-making processes.