Addressing the Radiology Bottleneck

Radiology departments face a widening gap between imaging demand and workforce capacity. Data from the Neiman Health Policy Institute indicates that outpatient imaging turnaround times more than doubled between 2014 and 2023. This delay creates a cascading effect in clinical settings; for instance, MIT Sloan research suggests a single CT scan can add up to 150 minutes to a patient's emergency department stay. Aidoc’s 'First Read' tool is designed to mitigate these delays by automating the initial drafting of radiology reports for chest radiographs, allowing radiologists to focus on verification rather than manual documentation.

Clinical Integration and Workflow

First Read is currently an investigational tool that has received FDA Breakthrough Device designation, a status intended to expedite the review of technologies that offer potential improvements for serious conditions. The tool is built upon Aidoc’s 'CARE' foundation model, which is trained on multimodal medical data.

Key operational features include:

  • Drafting Capability: The system can generate preliminary text for up to 100 pre-specified medical findings.
  • Human-in-the-Loop: The radiologist remains the final authority, responsible for reviewing, correcting, and signing every report.
  • Consistency: By automating the initial draft, the tool aims to reduce human error caused by fatigue during high-volume shifts or oversight of findings at the periphery of an image.
  • Continuous Monitoring: Aidoc plans to employ post-deployment monitoring to track model performance and identify potential drift, moving away from static validation toward real-world performance assessment.

The Shift Toward Foundation Models in Radiology

The development of First Read reflects a broader industry trend toward using foundation models to automate clinical documentation. Similar initiatives are underway across the sector, including Rad AI’s work on report impressions and Harrison.ai’s 'Harrison.Rad 1.5,' which incorporates clinical context and prior exams into report drafting. These tools represent a transition from single-purpose triage algorithms to broader, multimodal models capable of supporting complex, multi-task clinical workflows.