Enhancing TNM Staging in Breast Cancer: A Hybrid Approach with CNN, Edge Detection, and Self-Organizing Maps for Improved Accuracy

Abstract

Breast cancer remains a leading cause of mortality among women globally, underscoring the urgent need for improved diagnostic and staging techniques to enhance patient outcomes. This study aims to automate the TNM staging of breast cancer using a hybrid approach that integrates Convolutional Neural Networks (CNNs), edge detection methods, and Self-Organizing Maps (SOMs). Utilizing the Duke Breast Cancer MRI dataset, which provides detailed MRI scans crucial for accurate tumor characterization, the research addresses the limitations of traditional TNM staging, which often relies on manual interpretation by radiologists and can lead to inconsistencies and inaccuracies. Our approach combines CNNs with advanced edge detection algorithms and SOMs to automate and enhance the accuracy of breast cancer staging. The hybrid model effectively identifies and delineates tumor boundaries and critical anatomical features, offering a more reliable and objective evaluation. Notably, this method improves accuracy from 93% with conventional CNN models to 98%, representing a significant advancement in precision. This improvement not only provides more accurate diagnoses but also enables more personalized and effective treatment plans. For patients, this enhanced accuracy translates to better prognostic assessments and tailored treatments, potentially leading to improved outcomes and reduced likelihood of overtreatment or under treatment. For medical staff, the improved accuracy reduces the likelihood of misdiagnoses and enhances workflow efficiency by minimizing manual interpretation, thus alleviating some of the burdens associated with cancer staging. The model's performance is optimized through various testing methods and statistical evaluations, validating its stability and reliability. The integration of edge detection and SOMs captures comprehensive information, prevents overfitting, and provides valuable insights into data clustering. This combined approach supports personalized medicine by ensuring treatments are customized to individual patient characteristics, ultimately contributing to better survival rates and quality of life for patients.