An FDG-PET-Based Machine Learning Framework to Support Neurologic Decision-Making in Alzheimer Disease and Related...

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Distinguishing between neurodegenerative diseases is complex and requires specialized expertise.
• Clinical decision support systems (CDSSs) can aid diagnosis but face workflow integration challenges.
• A novel modeling framework using fluorodeoxyglucose PET (FDG-PET) imaging is proposed to overcome these challenges.

Purpose of the Study:

• To develop and evaluate a machine learning-based CDSS for diagnosing neurodegenerative diseases using FDG-PET imaging.
• To assess the performance and interpretability of the proposed framework.
• To demonstrate the potential for clinical integration of the developed tool.

Main Methods:

• A retrospective study utilizing FDG-PET images from 3,671 individuals across research studies and clinical patients.
• Development of the StateViewer framework, employing a neighbor matching algorithm to identify 9 neurodegenerative phenotypes.
• Evaluation through nested cross-validation, external validation on the Alzheimer's Disease Neuroimaging Initiative, and a radiologic reader study.

Main Results:

• The StateViewer framework achieved a sensitivity of 0.89 ± 0.03 and an AUC of 0.93 ± 0.02 in detecting 9 neurodegenerative phenotypes.
• In a reader study, clinicians using the StateViewer framework showed a 3.3 times greater likelihood of correct diagnosis compared to standard methods.
• The framework demonstrated high classification performance and interpretability.

Conclusions:

• The proposed FDG-PET imaging-based framework effectively aids in diagnosing neurodegenerative diseases.
• StateViewer addresses key challenges in integrating ML-based CDSS into clinical workflows.
• Further validation in diverse patient populations is warranted due to cohort limitations.