Why is amyloid important in Alzheimer’s disease?

Amyloid plaque accumulation in the brain is a defining biological feature of Alzheimer’s disease. Accurately identifying whether a person has abnormally high levels of amyloid, or is amyloid-positive, is critical for Alzheimer’s research and clinical trials. It directly affects who is eligible to participate in trials, how disease progression is monitored, and how treatment effects are measured.

Amyloid positron emission tomography (PET) imaging is widely used for this purpose. Expert visual interpretation of these scans is the regulatory gold standard, but scans close to the boundary between amyloid-negative and amyloid-positive can be difficult to interpret consistently. Reducing uncertainty in these borderline cases is important for improving trial efficiency and decision-making.

What did IXICO do?

IXICO assessed how three different approaches perform in determining amyloid status from PET scans:

• Expert visual reads performed by a radiologist
• Quantitative analysis using standardized uptake value ratio (SUVR) thresholds
• An IXICO deep learning artificial intelligence (AI) model trained to predict amyloid status from PET imaging data

SUVR provides a quantitative measure of amyloid burden by comparing tracer uptake in target brain regions with uptake in a reference region. In this work, a global cortical average (GCA) SUVR was used, which summarises amyloid uptake across large cortical regions. Amyloid status is then determined using a predefined threshold, with values above the threshold classified as amyloid-positive and values below as amyloid-negative.

IXICO utilised a large, well-characterised dataset ) to compare how closely SUVR-based and AI-based predictions aligned with expert visual reads. The AI approach was developed by IXICO, trained directly on PET images to predict amyloid status based on examples of expert visual assessment to learn complex spatial patterns of amyloid uptake that are not fully captured by global quantitative measures such as GCA SUVR.

What did IXICO find?

Both SUVR-based quantification and the AI model showed strong agreement with visual reads from a radiologist. Importantly, however, they tended to disagree with visual reads in different borderline cases. This suggests that SUVR and AI are sensitive to different features or patterns of amyloid uptake and may therefore provide additional complementary information.

Rather than replacing human interpretation, these methods have the potential to act as quantitative decision-support tools, helping radiologists interpret difficult scans with greater confidence and consistency.

Why does this matter?

This work highlights IXICO’s focus on improving the reliability and scalability of imaging-based biomarkers used in Alzheimer’s clinical trials. Even small improvements in classification confidence can have meaningful downstream impacts, including:

• More consistent participant selection
• Reduced variability across trial sites and readers
• Greater confidence in trial outcomes

As Alzheimer’s trials become larger and more complex, tools that support standardisation and operational efficiency are increasingly valuable.

What’s next?

This research represents an incremental but important step toward integrating AI-derived insights with established quantitative measures such as SUVR. IXICO’s ongoing work will further explore such approaches with validation across additional datasets and tracers to understanding how combined AI and quantitative metrics can best support visual interpretation in real-world clinical trial workflows.

Over time, this approach aims to strengthen IXICO’s platform by providing robust, data-driven tools that support high-quality decision-making in Alzheimer’s drug development.

Download here to view the full poster

At IXICO, we partner with biotech and pharmaceutical organizations to design and deliver multimodal biomarker and imaging strategies in Alzheimer’s disease clinical trials. To explore how we can support your programs with efficient plasma and imaging biomarkers, please visit our AD Content portal

 

 

 

 

 

 

---