AI analysis engine - diagnostics.ai

PCR.AI AI API: CE-IVDR Certified & MHRA registered AI engine with programmatic access

PCR AI API (AI Analysis Engine) - CE-IVDR & MHRA registered medical device

No user interface included - pure API access for integration

AI API Edition provides our CE-IVDR & MHRA registered AI analysis engine through clean API integration.

PARTNERSHIP APPROACHES

• Volume-based API model

• OEM integration model

• Long-term partnership agreements

DESIGNED FOR

• OEMs and test manufacturers

• Software developers

• Laboratory information system vendors

• Diagnostic equipment providers

• Instrument manufacturers

Please contact us for more details on intended usage and integrations.

Key Benefits

Pure AI power

Focus on the core analysis capabilities through clean API integration without interface components. Leverage our AI engine with >99.9% accuracy.

Regulatory Clarity

Benefit from our CE-IVDR certification & MHRA registration, reducing your regulatory burden.

Integration Flexibility

Implement our API according to your specific requirements with multiple integration options from RESTful API access to complete embedded solutions.

Developer friendly

Access comprehensive documentation and client libraries to accelerate your integration efforts.

Performance Optimised

Scale your analysis capabilities with our stateless, high-performance and high-throughput API designed for production environments.

Total Transparency

Clearly identify determined similarity between data-points to define positive, negative and review boundaries - both at training and at inferencing (routine analysis) analyses.

The AI Advantage

AI Outperforms Traditional Analysis Methods

Why not simply use thresholds and other built-in cycler methods?

The PCR.AI API is not reliant on thresholds or other single parameters of the curve but rather extracts multiple data-points for each curve entering the system, allowing the machine-learning to identify similar shaped curves in its training set for the assay (target).

This avoids issues such as those identified below.

Bundled Cycler Software: Just a fraction of the settings that can vary between runs and operators

Standardisation

End-users can change a wide-range of settings in common thermocycler software, which is perfect for researchers, but less helpful for routine clinical diagnostics where standardisation is key. However, curves can shift, settings can require adjustment to handle real-world issues, and experience in using such settings will vary between staff members.

Ideally thresholds would be standardised by SOPs and not changed between runs for the same targets, but this too can lead to artefacts (see below).

THE PCR.AI API HAS ZERO SETTINGS PER RUN - ONCE TRAINED (PER TARGET), RESULTS ARE 100% REPEATABLE.

False positives

Real-world example of sample incorrectly called positive by thermocycler using standardised thresholds

Linear rises above the threshold

PCR curves are normally assumed to be flat, except in cases of positive amplification. However, sometimes devices or assays misbehave, resulting in continual linear rises – even in negative controls (for example due to baseline drift).

THE PCR.AI API KNOWS THAT IT IS THE SHAPE OF THE CURVE THAT MATTERS, NOT ONLY THE HEIGHT.

Real-world example of samples incorrectly called positive by thermocycler using standardised thresholds.

Spikes (high noise)

Due to chemistry, instrument or user reasons, on occasion PCR curves exhibit high noise either persistently across cycles or in a single point in the run. In these circumstances, noisy negatives can pass thresholds – leading to false positive results.

THE PCR.AI API AVOIDS HIGH-NOISE (LOW SNR) FALSE POSITIVES.

False negatives

Example of incorrect thermocycler software CT being corrected in real-time by PCR.AI

Incorrect ‘above LOD’ or ‘Review’ results due to invalid thresholds or poor analysis by instrument software

The cycle threshold (CT) is key to PCR analysis however it is prone to being placed according to controls or other factors, which may cause individual samples to get incorrect results – for example a CT which is far too high or too low, leading to clinical errors. These issues can be particularly important in quantitative testing.

THE PCR.AI API DOES NOT USE THRESHOLDS TO DETERMINE CT; IT USES MORE RELIABLE PARAMETER EXTRACTION METHODS

Curves given similar CT by cycler software, but have very different shapes (same test and target, different runs)

Not all ‘late CTs’ are identical

Curves can have similar ‘cycle thresholds’ but vastly different shapes. Conventional methods cannot differentiate between these, and require manual review for non-standardised analysis.

THE PCR.AI API KNOWS THAT IT IS THE SHAPE OF THE CURVE THAT MATTERS, NOT ONLY THE CT.

MORE INFORMATION - Clinical References

NHS KCL - Potential pitfalls in analysing a SARS-CoV-2 RT-PCR assay and how to standardise data interpretation - clinical issues with the recent 'Relative Threshold' method from Thermo https://doi.org/10.1016/j.jviromet.2022.114589
NHS GLASGOW - Examples of clinical problems with threshold analyses link
OTHER ANALYSIS METHODS - coming soon!

PCR.AI Clinical Studies (NHS):

2024 NHS PCR.AI Quantitative Multiplex study - https://doi.org/10.1016/j.jviromet.2024.114981
2019 NHS PCR.AI Qualitative Multiplex study - https://doi.org/10.1016/j.jcv.2019.08.005