AI Automation in CT

Purpose	The typical workflow in CT consists of many manual steps. AI technology could streamline or perform many of these steps, which could free the technologist from the steps that are at risk for human-caused errors, and focus on the value-added steps that require higher level processing in order to mitigate errors.
Tag(s)
Panel	Technologist Facing
Define-AI ID	22140003
Originator	Samuel Brady
Lead	Samuel Brady
Panel Chair	Samuel Brady
Panel Reviewers	Technologist Facing
License	Creative Commons 4.0
Status	Public Comment

                               

Value Proposition

AI touchpoints at the CT scanner that help to reduce the possibility of human error that leads to excess radiation dose and/or repeat scanning.

Narrative(s)

In a busy imaging clinic, reducing the number of steps, or touchpoints, will provide a scenario for advanced patient safety. In most radiology clinics, Radiologists and CT technologists work together to verify orders for appropriateness and patient safety.  Areas for possible AI augmentation are highlighted for a standard patient examination. 

Category: patient safety. for CT examinations ordered with iodinated contrast, an AI algorithm will check the patient’s status and history to predict if the patient may have an allergic reaction, and make suggested modifications to the iodinated dosage if renal function results suggest impairment.  

Category: workflow efficiency. Once the patient is positioned on the scanner, and a localizer scan is performed, an AI algorithm will calculate the patient size and identify specific anatomy, body region and/or foreign objects within the body.  The scanner software will load the correct protocol based on the patient size and the ordered protocol.  If implants or other hardware are recognized in the body, temporary modifications to the acquisition technique factors can be suggested and advanced reconstruction algorithms (such as metal artifact reduction) can be added to the post processing workflow.  

Workflow Description

• Upon submission for a CT scan from an ordering physician, the CT order is reviewed for potential conflicts, previous scans, contrast contraindications, and/or suggested additional modifications to the type of scan ordered.  The examination is queued for protocol review with suggested amendments by an AI algorithm.   

• Patient is positioned/centered on the CT table and the CT localizer is acquired (normal, manual workflow step).

• An AI algorithm will identify the anatomy to be imaged based on the radiologist's protocol(e.g., if a general chest examination is ordered, the lungs will be identified in the localizer scan) and the appropriate beginning and ending scan range is set.  The Technologist will verify and adjust if necessary.

• The correct scan acquisition techniques (i.e., kV, mA, rotation time, etc.) will be loaded on the scanner console based on the patient’s size as measured from the localizer image.  The technologist will review for appropriateness and adjust as necessary.

• Following the completion of the transaxial scan acquisition, the scanner will review images for image quality, artifacts, and/or appropriateness then auto-process the multiplanar reformats. Additional reformats, such as metal artifact reduction, will be suggested or queued for technologist review.

The following considerations should be made for patient safety review:

1. Lab results and their findings (including patient history), as relating to kidney function or previous/known drug alergies, will be summarized and presented to ordering physician and radiologist(s). 

The following considerations should be made for gross anatomy segmentation/localization in a CT localizer scan: 

1. Different patient size, age,  and body shapes

2. Variety of clothing, jewelry/piercing variations (zippers, bras and other attenuating clothing accessories that may be visible in the localizer exam

3. Patients with differing implants and support equipment on or in the body (such as tubes, wires, leads, etc.)

The following considerations should be made for CT acquisition technique selection following the localizer scan: 

1. Different patient size, age, and body shapes

2. Type of ordered CT examination (this will determine radiation dose/ image noise levels in the reconstructed image); e.g., coronary plaque scoring scan of the heart and image of the brain for shunt positioning can each be low dose/high noise images since the diagnostic task is an inherently high contrast task that will not be limited by image noise/low radiation dose.

3. patient implant, surgery, etc. history

4. Iodinated contrast bolus timing

Inputs

Examination Order 

  

Primary Outputs

• Organ segmentation/classification; foreign object classification overlaid on the localizer scan 

• Suggested workflow modifications for imaging technique parameters, and post processing

Inclusion of CT contrast bolus timing may be developed based on patient specific lab results, physical presentation in the image, and or real-time data from an EKG.