Increasing the confidence of ¹⁸F-Florbetaben PET interpretations: Machine learning quantitative approximation

Area of Science:

• Neurology
• Nuclear Medicine
• Radiochemistry

Background:

• Alzheimer's disease and related dementias pose significant diagnostic challenges.
• Amyloid-beta (Aβ) positron emission tomography (PET) imaging is crucial for diagnosing these conditions.
• ¹⁸F-Florbetaben is a radiotracer used for PET imaging of Aβ.

Purpose of the Study:

• To evaluate the added value of semiquantitative parameters in ¹⁸F-Florbetaben PET imaging.
• To analyze patterns of ¹⁸F-Florbetaben brain deposition.
• To correlate semiquantitative data with visual assessment for improved diagnostic accuracy.

Main Methods:

• Retrospective multicenter analysis of 135 patients with mild cognitive impairment or dementia of uncertain origin.
• Visual interpretation of ¹⁸F-Florbetaben PET scans by two experienced observers.
• Semiquantification using standardized uptake value ratios (SUVRs) for cortical regions relative to reference regions.

Main Results:

• 72 out of 135 patients were classified as positive for Aβ deposition via visual assessment.
• Excellent interobserver agreement was achieved for visual interpretation.
• All SUVRs were significantly higher in visually positive scans, with prefrontal and posterior cingulate areas showing the strongest correlation with visual evaluation.
• ROC analysis indicated optimal diagnostic performance using SUVRs from specific target regions.

Conclusions:

• ¹⁸F-Florbetaben semiquantitative parameters enhance visual classification of Aβ deposition.
• The findings support the use of these parameters for machine learning applications in dementia diagnosis.
• Amyloid-beta deposition, while widespread, exhibits non-uniform and asymmetric distribution patterns across cortical regions.