A chest CT study does not appear on a radiologist's worklist instantly. Between the moment the scanner completes the acquisition and the moment the radiologist opens the study to read, a series of data handoffs and system interactions take place. Understanding that sequence helps clarify where AI triage fits, what it touches, and what it does not touch.
Step One: Acquisition at the Scanner
The pipeline begins when the CT scanner completes a study. Modern scanners reconstruct the raw detector data into a series of cross-sectional image slices, typically ranging from a few dozen to several hundred depending on the protocol and reconstruction settings. Chest CT studies commonly include multiple series: a standard axial reconstruction, coronal and sagittal reformats, and sometimes a high-resolution or thin-slice series for specific indications.
Scanner acquisition is governed by the protocol loaded at the scanner console, which specifies parameters like tube voltage (kVp), tube current (mAs), slice thickness, field of view, and reconstruction kernel. These parameters affect image quality and appearance, and they vary across scanner vendors and institutions. A chest CT from a 64-slice scanner running one vendor's protocol produces images with different characteristics than a chest CT from a 128-slice scanner running another vendor's protocol, even if the patient and clinical indication are identical. That variation is an important context for any AI model that will be processing these studies downstream.
Step Two: DICOM Transmission to PACS
Once the scanner completes reconstruction, it sends the image data to the PACS using the DICOM (Digital Imaging and Communications in Medicine) standard. DICOM is the universal format for medical imaging data: it carries not just the pixel data of each image slice but also a structured header containing patient demographics, study metadata, scanner parameters, and the ordering information that was pulled from the RIS at the time of scheduling.
The DICOM transmission happens automatically. The scanner and PACS communicate via a configured network connection; the scanner pushes the completed study and the PACS stores it. For most clinical sites, this transmission completes within one to five minutes of the scanner finishing reconstruction, though the timing depends on study size, network bandwidth, and PACS processing load.
From the PACS perspective, the study is now stored. It is associated with the patient record, the ordering physician, the modality, and the indication. But it has not yet been read, and it has not yet been evaluated for urgency.
Step Three: RIS Integration and Worklist Population
Alongside the PACS, most radiology departments run a RIS that manages the administrative lifecycle of imaging orders: order entry, scheduling, billing, and reporting. When a chest CT order is placed in the EHR, it flows to the RIS, which creates a study record and an accession number. When the scanner completes the study and sends it to PACS, the PACS uses that accession number to match the images to the correct RIS record.
This matching triggers worklist population. The study now appears in the reading queue for the relevant modality and body part. The radiologist assigned to chest CT reads sees the new case. The default sort key is arrival time: the new study lands at whatever queue position corresponds to when it arrived relative to the other pending studies.
The ordering indication, clinical notes, and referring physician information are all technically present in the study record at this point, but they are rarely structured in a way that the PACS worklist uses to determine read priority. The clinical context that might signal urgency exists somewhere in the data, but the worklist is not designed to act on it.
Step Four: The Wait in Queue
For many chest CT studies, the interval between DICOM arrival and radiologist open is measured in minutes during low-volume periods and in hours during high-volume periods, particularly overnight. A study that arrives at 2 AM during a wave of overnight inpatient scans might wait until 7 or 8 AM for a body radiologist to begin working through the morning queue.
This wait is the interval where triage-based reordering has its greatest effect. The study has arrived. The images are available. The radiologist simply has not yet reached it in queue order. If the study contains a time-sensitive finding, the wait is accumulating clinical risk without anyone in the care chain being aware of it, because the finding has not yet been seen.
Where AI Fits: The Pre-Read Triage Window
The triage window is the interval between DICOM arrival and radiologist open. It is the only interval in the pipeline where an AI layer can materially change the order of reads without disrupting any other clinical workflow.
ImageAssist operates in this window. When a chest CT study arrives in the PACS, a DICOM listener detects the new study and routes a copy of the image data to the ImageAssist analysis pipeline. The model processes the study, scores it for urgency across the finding types it monitors (hemorrhage, PE signal, high-grade nodules, effusion), and returns a decision: elevate or do not elevate. If the score exceeds the configured threshold, ImageAssist calls the worklist API to move the study to position one. The entire process completes in under 60 seconds from DICOM arrival in our pilot deployment experience.
Crucially, this operation does not modify the DICOM images, does not write to the patient record, and does not change the RIS order. It changes only the sort order of the study in the worklist. The radiologist who opens the study at position one sees exactly the same images they would see if the study had been opened at any other queue position. The AI layer has made a queueing decision, not a clinical one.
Step Five: The Radiologist Read
The radiologist opens the study in their PACS viewer and reads it. The read includes reviewing all series in the study, applying diagnostic reasoning, identifying and characterizing findings, and formulating an impression. For a chest CT that has been elevated to position one by ImageAssist, the radiologist may see a visual indicator noting the AI priority flag alongside the study in the worklist. That flag is informational: it tells the radiologist that the system identified a signal worth investigating, but it does not constrain or direct the read.
The radiologist confirms or does not confirm the finding. They issue the report in the RIS with their interpretation. If the finding warrants urgent communication, they initiate that communication per the department's critical results protocol. The AI layer's job ended when it moved the study to position one. Everything after that is the radiologist's clinical work.
Step Six: Report and Critical Results Communication
The radiologist's report is transcribed or dictated into the RIS, where it becomes part of the patient record and is available to the ordering physician via the EHR. Critical findings trigger a separate communication loop: a direct call to the ordering physician or the clinical team caring for the patient, documented in the report and the RIS. Some departments use automated critical results notification systems alongside the direct call; the specific protocol varies by institution.
The pipeline from scan to report is the same whether or not AI triage is involved. What AI triage changes is which study in the queue the radiologist reaches first. For the study that contains an urgent finding and was moved to position one, that change is the difference between a reading room that caught the finding as early as the clinical workflow permits and one that found it after the queue had run its natural course.
For a more detailed look at how the model that operates in the pre-read window is trained, see our article on machine learning for radiology triage.
