Hybrid AI systems in breast cancer histopathology classification: a systematic review and meta-analysis

Background Breast cancer (BC) diagnosis remains challenging in medical images as diagnostic interpretation varies among pathologists due to the complexity and variety of histopathological images, which can contribute to subjectivity and inconsistency in clinical decision-making. To address this ongoing issue and give pathologists better tools for detecting BC from histopathological data, researchers have been exploring advanced computer methods, like hybrid systems that combine machine learning and deep learning techniques. Methodology This systematic review was motivated by the need to assess the effectiveness, performance, data reliability, and real-world applicability of hybrid systems in histopathological image analysis of BC. Only studies on detecting BC that utilize hybrid artificial intelligence (AI) methods are considered, excluding simpler approaches or research focused on clinical data or other types of images apart from histopathological images. The studies were sourced from reputable databases such as PubMed, ScienceDirect, Scopus, IEEE Xplore Digital Library, MDPI, and PLOS One, and they span the period from January 2015 to April 2025. Data selection, measurement, reporting, and validation biases were assessed along with the overall bias in the 31 studies found according to the inclusion and exclusion criteria. Results The qualitative and meta-analysis on the 31 included studies showed that the pooled accuracy of these methods in multiclass and binary classification reached 90% and 94%, respectively. In terms of image data types, the pooled accuracy for whole slide images and region-cropped images was 86% and 96%, respectively. Conclusions Although there was a noticeable validation risk, as most of the researchers did not test their model on multiple datasets, these methods represent a promising pathologist AI assistant.