Machine learning methodologies in video capsule endoscopy-based bleeding analysis: A systematic review of progress and prospects (2008–2024)

The rising global mortality and morbidity of gastrointestinal (GI) bleeding, coupled with the limitations of traditional endoscopic methods, highlight the urgent need for innovative solutions in this field. Video Capsule Endoscopy (VCE) has emerged as a significant innovation, enabling a non-invasive visualization of the GI tract, crucial for identifying bleeding sources which may be inaccessible by traditional endoscopic methods. However, the efficacy of VCE is limited by challenges such as labor-intensive analysis and human error, where Machine Learning (ML) can offer significant improvements. This systematic review explores the role of ML in automating the analysis of GI bleeding in VCE, aiming to enhance diagnostic accuracy, reduce manual effort, and improve patient outcomes. A total of 114 studies published between 2008 and 2024 were systematically reviewed and categorized into four groups: segmentation, classification, detection, and combined methodologies. Each group was analyzed to assess the effectiveness and quality of the studies, using established assessment tools such as Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) and Prediction model Risk Of Bias Assessment Tool (PROBAST). An overview of open-source datasets, evaluation metrics, ML and its interpretability, and image processing methodologies used in the field was also conducted. Additionally, the review discusses the limitations and challenges encountered and highlights future prospects for ML methodologies in analyzing VCE frames for GI bleeding.