Machine-learning algorithms for detecting intracranial hemorrhage on head computed tomography

This is a protocol for a Cochrane Review (diagnostic). The objectives are as follows: The primary objective of this review is to determine the diagnostic accuracy of any form of ML algorithm for detecting ICH in participants who underwent hCT. We will assess diagnostic accuracy by individual and pooled indicators such as sensitivity and specificity with 95% confidence intervals, positive and negative likelihood ratios, and the summary receiver operating characteristic curve. Secondary objectives To investigate the following potential sources of heterogeneity in the diagnostic accuracy of ML: type of ML algorithm; year of development of the ML algorithm; certification of ML algorithm (Yes/No); type of reference standard; participant age (adults only/children only/mixed); hCT quality; hCT protocol; type of study (retrospective cohort study/prospective cohort study/RCT).