68/100

Moderate evidence

Cite
More
EmailLinkedInX

AI software lifted radiologists' accuracy on uncertain Parkinson-type MRI scans to 87%, study finds

Study title: Diagnostic accuracy and reliability of susceptibility map-weighted imaging in patients with clinically uncertain parkinsonian syndromes

Evidence type
Controlled or validated comparison
Sample
n = 180
Sources
1 sourcePeer-reviewed
Published
2 min read

Study title: Diagnostic accuracy and reliability of susceptibility map-weighted imaging in patients with clinically uncertain parkinsonian syndromes

A study of 180 patients with clinically uncertain parkinsonian syndrome tested whether AI-driven software could help radiologists read susceptibility map-weighted MRI scans. Radiologists alone reached a mean accuracy of 80%; with the software, that rose to 87%.

Read the study abstract

Parkinson's disease (PD) is characterized by degeneration of dopaminergic neurons in the substantia nigra, accompanied by increased iron accumulation, particularly in the nigrosome-1 subregion. Differentiating neurodegenerative disorders like PD from non-neurodegenerative causes of parkinsonism can be challenging. In clinically uncertain parkinsonian syndrome (CUPS), dopamine transporter (DAT)-single-photon emission computed tomography (SPECT) can aid diagnosis. However, susceptibility map-weighted imaging (SMWI…

Diagnostic accuracy and reliability of susceptibility map-weighted imaging in patients with clinically uncertain parkinsonian syndromes
"Doctor working in her office while using a computer and taking notes" by Shixart1985, CC BY 2.0.

The gain

In a study of 180 patients with clinically uncertain parkinsonism, four radiologists reading susceptibility MRI scans with AI software rose from 80% to 87% mean diagnostic accuracy, and agreement among them rose from a Fleiss κ of 0.65 to 0.86.

In brief

Agreement between the four radiologists also improved, from substantial (Fleiss κ 0.65) to almost perfect (0.86). The aim was to separate neurodegenerative parkinsonism from non-neurodegenerative causes, a distinction that is hard to make from symptoms alone.

The authors describe the MRI technique as an accessible alternative to dopamine transporter SPECT imaging, which is the usual aid in uncertain cases. The accuracy was measured against a final clinical diagnosis that included SPECT results.

Main points

  1. The study included 180 patients with clinically uncertain parkinsonian syndrome who had DAT-SPECT between January 2019 and July 2023.
  2. Of the patients, 84 had neurodegenerative and 96 had non-neurodegenerative parkinsonism; mean age was 67.9 and 120 were male.
  3. Four radiologists, blinded to symptoms and diagnosis, rated the nigrosome-1 region as normal, unclear or abnormal.
  4. Unaided, mean diagnostic accuracy was 80%, with individual radiologists ranging from 73% to 83%.
  5. With AI software, accuracy rose to 87%, with individual radiologists ranging from 85% to 91%.
  6. Inter-rater agreement increased from a Fleiss κ of 0.65 to 0.86 with the software.

The full story

Parkinson's disease involves loss of dopaminergic neurons in the substantia nigra and a build-up of iron, particularly in the nigrosome-1 subregion. Telling it from non-neurodegenerative causes of parkinsonism can be difficult, and DAT-SPECT is used when the clinical picture is unclear.

Four radiologists first rated axial susceptibility map-weighted images without knowing the patients' symptoms or diagnoses. They then re-read the scans with AI-driven software built to classify and quantify nigrosome-1 integrity. The reference standard was the final clinical diagnosis, which incorporated the DAT-SPECT results.

Accuracy ranged from 73% to 83% across radiologists alone and from 85% to 91% with the software. The narrower range and higher kappa indicate the software also made the readers' calls more consistent.

Sources

1 on file
  1. European Radiology Experimentaldoi:10.1186/s41747-026-00832-w
    Peer-reviewed2026-11-14

Talk it over

Three questions to start on, two more if you have time
Four radiologists went from 80% to 87% mean accuracy with the AI software, but the reference diagnosis already included DAT-SPECT results. How does that design shape what the gain can tell us?

Start with a question

AI software lifted radiologists' accuracy on uncertain Parkinson-type MRI scans to 87%, study findshttps://truvace.com/item/diagnostic-accuracy-and-reliability-of-susceptibility-map-weighted-imaging-in-pa

  1. Four radiologists went from 80% to 87% mean accuracy with the AI software, but the reference diagnosis already included DAT-SPECT results. How does that design shape what the gain can tell us?

    Answer in the thread
  2. Agreement among the radiologists rose from a Fleiss κ of 0.65 to 0.86. Why might readers agreeing more with each other matter, and when might it matter less than accuracy?

    Answer in the thread
  3. If you were one of the four radiologists who re-read the scans with the software after rating them alone, how would you decide when to trust its reading of the nigrosome-1 region over your own?

    Answer in the thread
If you have more time
  • Individual radiologists ranged from 73% to 83% alone and from 85% to 91% with the software. What would you want to know about who gained most, and what would that change about how the software is used?

    Answer in the thread
  • The study covered 180 patients who had DAT-SPECT between January 2019 and July 2023. What kind of follow-up study in a hospital you know would show whether the MRI technique can stand in for DAT-SPECT?

    Answer in the thread

What readers say