Deep Learning Techniques for Subtype Classification and Prognosis in Breast Cancer Genomics: A Systematic Review and Meta-Analysis

Abstract

Breast cancer is a significant global health concern, characterized by its complex and heterogeneous nature, which presents challenges for accurate diagnosis and effective treatment. Traditional classification methods for breast cancer subtypes and prognosis prediction often lack precision. In contrast, recent advances in deep learning have shown great potential to enhance diagnostic accuracy and improve patient outcomes by leveraging complex genomic data. This systematic review and meta-analysis aim to evaluate the effectiveness of deep learning models in classifying breast cancer subtypes and predicting prognosis. By focusing on studies published between January 2013 and February 2024, sourced from Scopus and PubMed databases, this review analyzes the performance of models such as Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs). The results from 221 studies highlight that deep learning models significantly outperform traditional methods, achieving an average AUC of 0.893 and accuracy rates between 65.92% and 93%. These models demonstrate their ability to detect subtle genomic patterns associated with disease progression and patient outcomes, marking a substantial advancement in personalized medicine and breast cancer diagnostics.