Detecting Image Quality in Medical Imaging

Purpose

Image quality is crucial for proper diagnosis from a medical imaging study. Most institutions have ongoing quality improvement procedures and policies to improve and maintain diagnostically acceptable image quality. However, in a busy clinical environment, often times suboptimal images are acquired, and it is impractical for every radiologist to flag every case and to follow up to resolution in an increasingly large, corporate-like healthcare practice setting. If a computer algorithm can detect suboptimal study automatically, it will make it easier for quality improvement committees to review and resolve issues. Furthermore, if the algorithm can detect image quality in real time, it can warn technologists of suboptimal study before patient leaves and study can be re-done in time.

Tag(s)

Non-Interpretative

Panel

Technologist Facing

Define-AI ID

19140001

Originator

Stephanie Jo, MD, PhD, CIIP
Lead Stephanie Jo, MD, PhD, CIIP

Panel Chair

Ross Filice
Non-Interpretative Panel Chairs Alexander J Towbin, Adam Prate

Panel Reviewers

Technologist Facing

License

Creative Commons 4.0 
Status Public Commenting 
                               

Clinical Implementation


Value Proposition


Suboptimal image quality interferes with image interpretation, leading to incorrect or uncertain diagnosis, as well as increased interpretative time, and potentially suboptimal patient care. Non-diagnostic quality image may require repeating the study or may trigger a recommendation of a repeat exam or different study modality (e.g., CT or MRI), adding to healthcare cost and inconveniencing the patient. Automatic detection of image quality allows for correction of the problem at the time of the imaging exam or shortly thereafter and could result in better diagnosis and patient care while reducing repeat imaging.

Narrative(s)


Obese patient comes to the emergency room after a fall due to knee pain. Adjusting kVpand mAs can achieve reasonable image quality for diagnosis of fracture on a radiograph. Standard kVpand mAs may not be sufficient for an obese patient due to increased scatter. However, the emergency room is really busy, with multiple studies waiting to be done, the patient is in pain and has difficulty getting into proper position, given all the distractions standard settings are used. The radiologist views the study, notices poor quality, and gives the disclaimer of “cannot rule out fracture.” Emergency room doctor then orders a CT to evaluate for fracture. If the algorithm can detect these cases retrospectively, they can be reviewed by a quality improvement committee and can be remedied by procedures & policies (e.g., for BMI > 30, use adjusted settings). If the algorithm can catch these cases in real-time before the technologist completes the study, additional images can be obtained while the patient is still in the X-ray suite, and subsequent CT may not need to be performed.

Patient with hip replacement hardware presents for CT for abdominal pain. The patient forgets to mention the hardware, as it is in the hip and pain is abdominal. The technologist and radiologist do not have this information as the referring physician is outside the system. The referring physician does not indicate in the CT abdomen/pelvis order that the patient has hip replacement hardware. Thepatient gets a CT of the abdomen/pelvis without metal suppression protocol. There are marked metal artifacts obscuring the pelvis, limiting evaluation. Radiologist cannot see structures such as bladder, reproductive organs, , and pelvic small bowel. Patient needs to return for a repeat study, or the clinician needs to make decision based on the low-quality study. Clinician may order another study (e.g. endoscopic ultrasound, GI fluoroscopic study) for evaluation of unseen structures. Another possibility is that the radiologist may mistake an artifact as an abnormality, and may recommend MRI or other additional imaging studies for follow up. If the algorithm can catch these cases retrospectively, they can be reviewed by a quality improvement committee and can be remedied by procedure & policies (strict review of all hardware before CT, use of scout image to guide CT protocol selection for hardware). If the algorithm can catch these cases in real-time before the technologist completes the study, additional images can be acquired with the proper metal suppression protocol while the patient is still in the CT suite.

Workflow Description



  • Patient with hip replacement arrives in ED with abdominal pain

  • CT acquired using routine protocol (no metal suppression)

  • Algorithm analyzes scout images; detects presence of metal implants; flags / notice technologist of choosing alternative imaging protocol with metal artifact reduction algorithms

  • Technologist receives alert; realizes that hip hardware is present; obtains new images through the pelvis with metal suppression

  • All images are tracked for review; radiologist is able to make accurate diagnosis with high quality images; no repeat imaging necessary


Considerations for Dataset Development



Motion artifact (CT, DR, MR):

present, absent

Hardware artifact (CT, MR):

present, absent

Suboptimal beam penetration due to body habitus (DR, CT):

present absent

Suboptimal FOV for indication (CT, MR, DR):

present, absent

ring artifacts (CT):

present, absent

Shading (CT):

present, absent

cupping/capping (CT):

present, absent

Patient anatomy outside the field of view(CT):


present, absent

Clinical note: causes shading and streaks

Technical Specifications


Inputs

DICOM Images

Procedure

All Imaging

Views

All

Data Type

DICOM

Modality

X-ray, CT, MRI




Primary Outputs


Hardware Artifact

RadElement ID


Definition

Presence of hardware artifact

Data Type

Categorical

Value Set

0- Hardware artifact absent

1- Hardware artifact present

2- Unknown

Units

N/A


Motion Artifact

RadElement ID


Definition

Presence of motion artifact

Data Type

Categorical

Value Set

0- Motion artifact absent

1- Motion artifact present

2- Unknown

Units

N/A



Additional Imaging Recommendation

RadElement ID


Definition

Recommendation for additional imaging based on prior image adequacy

Data Type

Categorical

Value Set

0- Imaging adequate

1- Imaging inadequate; further imaging recommended

2- Unknown

Units

N/A

Future Development Ideas


  • Expansion into other more uncommon image quality problems (MRI pulsation artifact, RF leak, other MR-specific problems; CT contrast timing problems; DR positioning problems i.e. patient is oblique when should be AP etc).