JMIR Publications

ABSTRACT

Background:

Although DICOM metadata is widely used to manage medical imaging data and support clinical workflows, its suitability as a sole basis for automatic CT series labeling and characterization is limited. DICOM metadata are frequently inconsistently populated or institution-specific, including the use of unregulated private tags and variable reliability even within standardized fields. Consequently, automated series selection for downstream AI applications often remains unreliable, necessitating manual curation within clinical workflows.

Objective:

This study presents Orchestrate, a modular AI framework for fully automated orchestration of CT imaging data. By integrating a hierarchy of deep learning models, Orchestrate enables pixel-level classification and routing of CT series and accurate metadata-independent identification of anatomical regions, contrast-enhanced series, and reconstruction kernels, supporting seamless downstream AI integration without manual curation.

Methods:

Orchestrate was developed using YOLOv8-based models and classifiers to detect anatomical regions, classify contrast-enhancement, recognize reconstruction kernels, and infer laterality, leveraging 114,627 CT studies. Its performance was evaluated on 745 internal and 70 external CT studies, using DICOM metadata and expert-annotated ground truth. Clinical utility was assessed through cohort selection tasks involving three predefined target cases, with expert review of the selection accuracy.

Results:

Across all studies, DICOM metadata revealed frequent inconsistencies, including missing kernels (9%) and contrast media information (17%). In contrast, Orchestrate achieved high classification performance across internal (average F1 score of 0.970, accuracy of 99.15%) and external cohorts (average F1 score of 0.992, accuracy of 95.64%). The three clinical use cases of liver, thoracic, and whole-body scans demonstrated an overall selection accuracy of 99.68%, indicating that the framework reliably identifies and routes appropriate CT series.

Conclusions:

Orchestrate enables fully automated pixel-based classification, detection, and semantic description of CT series, reducing reliance on manual selection and the risk of inconsistent metadata. By generating standardized semantic content, the framework improves interoperability with clinical systems and supports a reliable, reproducible integration of AI-driven imaging pipelines into clinical workflows.