An Automated Hybrid Deep Learning-based Model for Breast Cancer Detection using Mammographic Images

Breast cancer is a disease in which abnormal breast cells grow uncontrollably and develop into a tumor. It is one of the most common cancers that affects women around the world. The most often used imaging tool for identifying breast cancer is mammography. Early and accurate identification of tumors can be crucial for effective treatment and planning, which reduces mortality rates. This work proposes a novel hybrid deep learning-based automated framework for early and accurate breast cancer detection using mammographic images.

The methodology integrates multiple components into a hybrid model. Initially, mammographic images from the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) undergo a preprocessing step, uses a guided filter to remove noise and enhance the visibility of regions of interest (ROI). Modified Dingo Optimization (MDO) algorithm is used to segment the tumour-affected regions, not only to identify abnormalities localized in a single region of the breast but also to effectively detect multiple abnormal areas distributed across different tissue regions. Deep features are then extracted using a pretrained U-Net architecture. The Search and Rescue Optimization (SRO) algorithm was utilized for feature optimization to select the most relevant deep features, reducing dimensionality and enhancing the model's diagnostic accuracy. A Dual Stage Spiking Convolutional Neural Network (DSS-CNN) is implemented for classification and enhancing the model's ability.

The proposed hybrid deep learning model achieves outstanding performance, with an accuracy of 98.598%, precision of 97.343%, recall of 97.514%, and an F-measure of 96.89%. Comparative analysis confirms that the approach significantly reduces false positive and false negative rates, outperforming existing state-of-the-art techniques.

The proposed robust end-to-end system for early and accurate breast cancer detection demonstrates the efficacy of the framework in improving diagnostic accuracy, precision, recall, and F-measure, offering valuable support in clinical decision-making.