The use of AI in imaging for Alzheimer’s Disease
October 2024
The AD landscape is complex but evolving with the recent approval of amyloid-targeting therapies (ATTs), which have enabled an economy built around supporting the new unmet needs created by these therapies
Barriers to address for improving patient outcomes
Diagnosis
- PCPs are not set up for success to provide adequate workup
- Current cognitive and functional tests have limitations and require repeat testing to detect decline over time
- Blood biomarkers (BBMs) can address some of these limitations but are yet to be FDA-approved
- The use of AI technology can help accelerate image processing and interpretation, aiding with diagnosis
Integration
- AI-powered software that can seamlessly integrate with MRI and PET scan devices could alleviate long waiting time
- Hiring and training required personnel needed to carry out all required steps for diagnosis and medication administration (eg, there is a need for more radiologists as current imaging centers have bandwidth issues)
- There currently appear to be no major partnerships between companies providing AI-powered software and care systems
On this page, we will provide an overview on the use of AI-powered software for MRI and PET scan imaging interpretation, including the landscape, available tests, and industry trends
EXECUTIVE SUMMARY – THE AI LANDSCAPE SPACE TODAY
AI-powered software tackling some of the main unmet need have already been developed and are currently awaiting FDA 510(k) approval
Current AI-powered software are designed for a broader usein neurology
- Current AI-powered software have been developed for not only Alzheimer’s disease but for a larger number of neurodegenerative diseases, such as looking at changes in white and grey matter and volumetric changes
- The approval of ATTs has created a new set of needs and challenges that current approved products do not address, like the need for centiloid count on PET scans
Adding functionalities such as ARIA monitoring can help address some of the unmet needs created by ATTs
- Software addressing some of the unique unmet needs like detecting and monitoring for ARIA are currently awaiting FDA 510(k) approval and could be granted as early as Q1 2025
- Different pricing models, such as subscription-based models or pay-as-you-go models, could help with the adoption of these technologies in large and small healthcare systems
Did you know?
ARIA is one of the side effects of monoclonal antibodies directed to amyloid beta aggregates such as lecanemab and donanemab
EXECUTIVE SUMMARY – THE FUTURE AI IMAGING LANDSCAPE
Ease of integration and additional features that can aid with image interpretation and diagnosis will be key moving forward
Unmet need: Testing for multiple targets yields the most promising outcomes
Future trends: Ability to detect ARIA-E and ARIA-H events and additional features, such as the ability to detect the potential risk of developing ARIA, could help improve the safety of ATT
Unmet need: Most companies currently only offer one payment system, although some have shifted to provide multiple options
Future trends: Having different pricing models for different healthcare systems could help those with large or small volume of patients
Unmet need: The need to integrate with a cloud-based AI could help with faster computation and analysis but have the risk of not being available if there are connectivity issues
Future trends: Products that don’t over-rely on an internet connection for some of their key features could bypass this issue, although computational analysis could be compromised
There is great potential for the use of AI in helping neurologists and radiologists with diagnosis and monitoring for safety, yet FDA approval is critical for their adoption in the real world
Detailed Report
AI imaging could be used alongside BBMs to reduce the need for CSF testing in AD; however, it is unlikely to replace CSF testing as a standalone modality in the near-term
Drivers of AI-Driven Imaging Adoption
Driver
Decreased Scan Time
Description
Studies have shown up to 4x faster scan times with AI processing tools1.
Stakeholder Benefit*
Increased machine availability would allow imaging centers to scan a greater volume of patients
*Benefit: Gain to stakeholder resulting from increased MRI/PET usage
Driver
Increased Confidence in Scan Quality
Description
Real-time motion tracking and gating optimization, and post-processing AI tools could correct for motion artifact in MRI and PET scans
Stakeholder Benefit*
Greater scan quality would reduce the need for repeat scans and patient callbacks, minimizing unnecessary strain on imaging facilities
Movement correction would allow for roomier, less restrictive scanners, reducing patient discomfort
Driver
Enhancement of Weaker Scanners
Description
Improved accuracy and resolution may enable 1.5T magnetic field strength scanners to approach the quality of 3T scanners, which are 30%-40% more expensive2,3.
Stakeholder Benefit*
A broadened network of capable scanners would improve patient access to minimally-invasive diagnostics.
1NYU Langone Health NewsHub; 2Siemens Healthineers; 3LBN Medical
Critical Implication Given current capabilities, AI-driven imaging is expected to be used alongside BBMs in the near term
With an increase in AI-powered software, the ability to integrate with imaging devices, unique software features, and pricing models will help differentiate currently available options for AD
Integration
Most software offers the ability to integrate with Picture Archiving and Communication Systems (PACS), but not all are accompanied by additional stand-alone/third-party or web-based platforms that could help differentiate themselves from the competition by offering seamless user experiences
Software features
Key features like the ability to detect ARIA in patients undergoing ATT, volumetric measurements of white or grey matter, or to perform contrast weighted images, or multi-time-point or normative comparisons will also determine which software can offer the best set of features that respond to healthcare system needs
Pricing models
Accommodating pricing models can be matched to needs of institution type; for example, subscription based for large institutions or pay as you go for smaller radiology centers
With an increase in AI-powered software, the ability to integrate with imaging devices, unique software features, and pricing models will help differentiate currently available options for AD
Product | Company | Integration | Key features |
---|---|---|---|
Pixyl.Neuro.BV1 | Pixyl | Integration in standard reading environment (PACS), integration via AI marketplace or distribution platform, stand-alone web based | Brain volume quantification, brain segmentation, comparison with normative values, longitudinal analysis |
ARIAscore structure2 | ARIAMed† | Integration in standard reading environment (PACS) | Brain tissue and anatomy segmentation, volume quantification, normative comparison, report generation, WMH detection and quantification |
QyScore3 | Qynapse | Integration in standard reading environment (PACS), integration via AI marketplace or distribution platform, stand-alone web-based | Automatic labeling and volumetric quantification of segmented central nervous system structures; decreases image reading variability and segmentation errors |
Trace4AD4 | DeepTrace Technologies | Stand-alone third-party application, stand-alone web-based | Provides risk (low or high) of having or progressing to AD within 24 months by an automatic reading of the subject’s brain grey matter obtained from a 3D structural T1-weighted MRI brain study, also in combination with subject’s neuropsychological measures |
Product Capabilities
Product | Detection/diagnosis | Prognosis | Monitoring | ARIA detection |
---|---|---|---|---|
Pixyl.Neuro.BV1 | ||||
ARIAscore structure2 | ||||
QyScore3 | ||||
Trace4AD4 |
FDA Clearance and Pricing
Product | Company | FDA clearance date | Key features | Pricing model |
---|---|---|---|---|
SubtlePET1 | Subtle Medical | 12/5/2018 | SubtlePET image processing software reduces noise to increase image quality using a deep neural network-based algorithm; “denoises images conducted in 25% of the original scan duration” | Unknown |
Neurocloud PET2 | Qubiotech | Not yet (CE certified, Class I) | Identify and quantify regions with abnormal metabolism, positive/negative amyloid result, customizable report | Pay-per-use, Subscription, Customizable Plans |
Neurophet SCALE PET3 | Neurophet | 08/05/2022 | Quantifies SUVR of biomarkers (e.g., amyloid, tau) targeted by various radiotracers using PET images; Accurately measures atrophy of white matter and grey matter caused by neurodegenerative disorders to provide analysis results and SUVR calculation for 91 brain regions | Subscription, one-off payment; based on pay-per-scan |
PalRe™4 | PAIRE | Unknown | Supports decision-making by detecting and segmenting lesions on PET scans to extract features that require attention | Unknown |
Product | Noise reduction | Detection/diagnosis | Monitoring |
---|---|---|---|
SubtlePET1 | |||
Neurocloud PET2 | |||
Neurophet SCALE PET3 | |||
PalRe™4 |
Detection and Monitoring
Product | Key features | Detection/diagnosis | Monitoring |
---|---|---|---|
Neurocloud PET3 | Identify and quantify regions with abnormal metabolism, positive/negative amyloid result, customizable report | ||
Neurophet SCALE PET4 | Quantifies SUVR of biomarkers (e.g., amyloid, tau) targeted by various radiotracers using PET images; Accurately measures atrophy of white matter and grey matter caused by neurodegenerative disorders to provide analysis results and SUVR calculation for 91 brain regions |
ARIA Detection Capabilities
Product | Key features | ARIA detection |
---|---|---|
Neurophet AQUA AD1 | Brain region segmentation, volume quantification, normative comparison, report generation, white matter hyperintensity quantification, multi-time-point analysis, ARIA monitoring | |
icobrain aria2 | Automated quantification of ARIA-E and ARIA-H |