Radiology artificial intelligence, a systematic evaluation of methods (RAISE): a systematic review protocol.

No comprehensive synthesis currently exists to map the diverse landscape of machine learning applications within medical imaging. Prior research has shown rapid growth in computer vision tools for diagnostic tasks. That uncertainty drove the need for a standardized evaluation of study quality and design. It was already known that existing guidelines focus primarily on ethics and data management. This gap motivated a formal assessment of the entire domain. Previous investigations often lacked a unified framework for comparing disparate clinical questions. Researchers have struggled to identify clear trends in methodological approaches across different subspecialties. This systematic review protocol addresses these limitations by establishing a rigorous, task-centered analytical structure.

The aim of this study is to investigate the use of artificial intelligence as applied to clinical radiology. This systematic review seeks to identify the specific clinical questions currently being asked by researchers. The team intends to clarify which methodological approaches are most frequently applied to these questions. By analyzing trends over time, the authors hope to provide a clearer picture of the field's evolution. The researchers also want to assess the quality of study designs currently prevalent in the literature. This effort addresses the lack of comprehensive reviews covering the entire domain of medical imaging informatics. The protocol provides a structured way to compare disparate studies using a task-centered approach. Ultimately, the study seeks to establish a baseline for understanding how these technologies are implemented in real-world clinical settings.

The review approach follows the Preferred Reporting Items for Systematic Review and Meta-Analysis guidelines for transparency. Researchers will conduct searches across MEDLINE and EMBASE databases to identify relevant clinical trials. The team applies no language restrictions to ensure a global perspective on the subject matter. Data extraction will focus on specific trial characteristics to facilitate a detailed narrative synthesis. A task-centered strategy replaces traditional modality or subspecialty groupings to improve analytical focus. Sub-group analysis will categorize findings into segmentation, identification, classification, and regression or prediction tasks. A separate assessment will specifically examine data related to pediatric patient populations. Ethical approval remains unnecessary because the study relies exclusively on publicly accessible information from existing clinical trials.

Key findings from the literature will emerge from a structured analysis of existing clinical trials. The protocol establishes a clear registration number, PROSPERO CRD42020154790, for tracking the study progress. Researchers will identify the primary clinical questions currently driving innovation in the field. The team will quantify trends in computational usage over time to highlight shifts in research priorities. Methodological approaches will be evaluated to determine their prevalence across different diagnostic tasks. The narrative synthesis will provide a comprehensive overview of study designs and quality benchmarks. Data extraction will capture specific trial characteristics for each identified study. The final results will offer a detailed map of how machine learning is currently integrated into diagnostic workflows.

The authors propose that this protocol will clarify the current state of machine learning in diagnostic imaging. This synthesis aims to identify specific clinical questions addressed by existing literature. The researchers intend to map methodological trends over time to guide future inquiry. By categorizing tasks, the team expects to reveal patterns in how algorithms are applied. The review will provide a comprehensive overview of study designs and quality metrics. The authors suggest that focusing on tasks rather than modalities will improve clarity. This work will offer insights into the application of predictive models in pediatric populations. The final publication will serve as a resource for standardizing future research practices in the field.