Parallel convolutional SpinalNet: A hybrid deep learning approach for breast cancer detection using mammogram images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer remains a leading cause of mortality in women, emphasizing the need for early and accurate diagnostic methods.
• Timely detection significantly improves patient outcomes and survival rates, making advanced imaging analysis crucial.

Purpose of the Study:

• To propose and evaluate a novel deep learning model, Parallel Convolutional SpinalNet (PConv-SpinalNet), for efficient and accurate breast cancer detection using mammogram images.
• To enhance the diagnostic performance by integrating various image processing and feature extraction techniques.

Main Methods:

• Mammogram images were pre-processed using Gabor filters and tumors were segmented with LadderNet.
• Advanced augmentation techniques (Image manipulation, erasing, mix) were applied to segmented samples.
• A comprehensive feature extraction phase included CNN features, Texton, Local Gabor Binary Patterns (LGBP), Scale-Invariant Feature Transform (SIFT), and Local Monotonic Pattern (LMP) with Discrete Cosine Transform (DCT).
• The PConv-SpinalNet, combining Parallel Convolutional Neural Networks (PCNN) and SpinalNet, was utilized for final detection.

Main Results:

• The PConv-SpinalNet model demonstrated a high accuracy of 88.5%.
• Key performance metrics included a True Positive Rate (TPR) of 89.7%, True Negative Rate (TNR) of 90.7%, Positive Predictive Value (PPV) of 91.3%, and Negative Predictive Value (NPV) of 92.5%.

Conclusions:

• The proposed PConv-SpinalNet model shows significant potential for improving the accuracy and efficiency of breast cancer detection in mammography.
• This deep learning approach offers a promising tool for early diagnosis, potentially reducing mortality rates and enhancing patient care.