Volume 10 Issue 2

Breast Cancer Prediction Using Machine Learning Models and Digital Pathology: A Systematic Review

Kehinde

Abstract

Background: Breast cancer (BC) remains a paramount global health challenge, driving the oncology community toward more precise and efficient diagnostic methodologies. The advent of digital pathology has been transformative, creating an unprecedented opportunity to apply computational intelligence to the analysis of tissue samples. In this context, a diverse array of artificial intelligence (AI) techniques including traditional machine learning (ML), data-intensive deep learning (DL), and integrated hybrid learning (HL) models are being actively developed for BC prediction. However, the rapid proliferation of research in this domain has led to a fragmented understanding of their comparative strengths and practical implementation barriers. This systematic review was therefore undertaken to synthesize the existing evidence, critically appraising the distinct capabilities and limitations of ML, DL, and HL in the analysis of digital pathology images. Aim: This study aims to systematically review and analyze the application of ML, DL and HL techniques in BC prediction using digital pathology, highlighting their comparative strengths, limitations, and impact on diagnostic accuracy. Method: Following Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines, a systematic review was conducted using Google Scholar to identify relevant studies. Search terms were formulated to retrieve literature on ML and DL models in digital pathology. Articles were screened based on predefined inclusion and exclusion criteria. Result: A total of 107 articles were reviewed, including 19 on ML models, 76 on DL models, and 12 articles on HL models. The results highlight the strengths and limitations of each model technique, with DL model being the most widely used approach. Conclusion: The assessment of ML, DL, and HL approaches reveals their individual capabilities and shortcomings in predicting BC via digital pathology. DL is the current front-runner, excelling with large image sets, but the slower uptake of ML and HL methods shows room for exploration. This is particularly true for needs like model transparency, combining diverse health records, and ensuring robustness across populations. The conclusions also call for more international partnerships and greater involvement from overlooked areas, especially African nations, to make certain that innovations in computational pathology are universally applicable and address a wide range of medical environments.

Keywords

Full Text

Download

References