Deep-learning-based synthesis of PET maps from MRI for Alzheimer’s disease detection

This technology is a deep-learning-based model that synthesizes positron emission tomography (PET) maps of pathological burden from MRI to detect Alzheimer’s disease.

Unmet Need: Safe, accessible methods for PET imaging for Alzheimer’s disease detection

Positron emission tomography (PET) is considered the gold standard molecular measure of Alzheimer’s pathology. However, this method is expensive, not universally available, and exposes patients to ionizing radiation. In contrast, T1-weighted (T1w) MRI is widely available, cost-effective, and does not expose patients to ionizing radiation. However, T1w MRI is unable to generate standardized uptake value ratio (SUVR) maps of pathological burden, which is necessary for determining the molecular pathology of Alzheimer’s disease.

The Technology: Deep-learning-based model for generating synthetic PET maps from MRI

This technology is a deep-learning-based framework that generates synthetic PET maps from T1w MRI for the detection of Alzheimer’s disease. Structural, morphometric, and vascular features are integrated from MRI to synthesize PET SUVR maps. This hybrid deep learning architecture achieves high concordance with ground-truth PET imaging and improves performance when vascular and morphometric priors are included.

Applications:

• Diagnostic tool for Alzheimer’s disease
• Early risk stratification
• Longitudinal disease monitoring
• Therapeutic response tool

Advantages:

• Cost-effective
• Non-invasive and non-ionizing
• Model input derived from widely available MRI medical equipment
• High concordance with ground-truth PET imaging
• Improved performance with vascular and morphometric priors

Lead Inventor:

Jia Guo, Ph.D.

Related Publications:

Tech Ventures Reference:

• IR CU26244

• Licensing Contact: Jerry Kokoshka