Technology

The only AI that speaks MRS.

We didn’t adapt a general-purpose AI for MRS.We built an AI engine from MRS physics — from the ground up.

Core Engine

Not a black box. An AI that obeys physics.

An MRS spectrum is a chemical fingerprint — each peak shaped by the quantum-mechanical properties of the molecule it represents. These aren’t patterns to guess at. They’re laws to build on.

METLiT’s deep learning doesn’t just learn from data. It searches for answers within the boundaries that MRS physics allows. The result: an engine that is accurate not by coincidence, but by design.

1

Data Efficient

Learns from physics, not just volume. Superior generalization even with limited clinical datasets.

2

Physically Consistent

The model cannot produce results that violate spectroscopic principles. This is not a post-hoc filter — it is built into the architecture.

3

Transfer-Stable

Same physics, same performance. Consistent results across institutions, scanners, and protocols.

Input — Raw MRS Spectrum
4.03.53.02.52.01.51.0Chemical Shift (ppm)

Overlapping metabolite signals — unresolvable by conventional methods

Physics-Informed Deep Learning Engine
1×13×35×5PoolInc-ARed-BInc-C
×TMonte Carlo Dropout
MRS physics constraints
17 Metabolites
NAA
Cho
mI
Glu
Gln
GABA
GSH
Lac
+ 9 more metabolites
Uncertainty
NAA/Cr
2.7%
Cho/Cr
3.9%
mI/Cr
6.2%
Glu/Cr
10.8%
Gln/Cr
11.5%
GABA/Cr
8.5%
GSH/Cr
9.3%
Lac/Cr
7.2%

Bayesian uncertainty per metabolite

Physics-informed decomposition — not curve fitting

Performance

Side by side, there is no comparison.

Aspect
Conventional (LCModel, etc.)
METLiT
Brain Metabolites
~5
17
Liver Metabolites
Fat ratio only
Fat + multiple metabolites
Analysis Time
Tens of minutes to hours
< 15 seconds, fully automated
Expert Dependency
Ph.D.-level specialist required
No specialist needed
Scanner Compatibility
Vendor-dependent
GE · Philips · Siemens Healthineers
System Integration
Manual data transfer
Cloud API + PACS auto-sync
Confidence Reporting
None
Bayesian uncertainty quantification

Reliability

Every number comes with a confidence score.

Most AI gives you an answer. It doesn’t tell you how much to trust it. METLiT applies Bayesian deep learning to quantify uncertainty for every single metabolite. High confidence? Act on it. Low confidence? We tell you before you have to ask.

NAA/Cr
1.26 ± 0.04
2.7%
Cho/Cr
0.21 ± 0.02
3.9%
mI/Cr
0.89 ± 0.06
6.2%
Glu/Cr
1.14 ± 0.09
10.8%
GABA/Cr
0.17 ± 0.03
8.5%
GSH/Cr
0.20 ± 0.04
9.3%
Uncertainty range
Quantified value

High Confidence

Clinically actionable — proceed with decision-making.

Moderate Confidence

Reliable with context — specialist review recommended.

Data Quality Issue

Signal quality insufficient — rescan recommended.

Universality

One AI engine. Every major MRI scanner.

Clinical MRI scanners vary — by vendor, by institution, by configuration. Conventional MRS software breaks when the setup changes. METLiT was built to handle this variability from day one. GE, Philips Healthcare, Siemens Healthineers — same AI, same result quality.

GE Healthcare

Nearly all versions of P-file format fully compatible

Fully Verified

Philips Healthcare

SPAR / SDAT and Enhanced DICOM compatible

Fully Verified

Siemens Healthineers

.rda, TWIX, and Enhanced DICOM compatible

Fully Verified
One AI engine — all three vendors

Automation

From scan to report. No manual steps in between.

MRS data analysis has always been a multi-stage expert workflow: signal processing, quality control, quantification, interpretation, reporting. Each stage traditionally required different specialists.

METLiT automates the entire pipeline end-to-end. Where autonomous decision-making is needed — signal quality assessment, clinical pattern recognition — dedicated AI agents handle the reasoning. Where deterministic precision is needed — quantification, data ingestion, report generation — purpose-built engines execute flawlessly.

Conventional Approach
Raw Data
Manual — specialist required at every step
Preprocessing
Ph.D. specialist
Quality Check
Ph.D. specialist
Quantification
Ph.D. specialist
Result
Separate interpreter needed
Every step requires a different specialist — slow, expensive, inconsistent
METLiT End-to-End AI Platform
Raw Data
Fully Automated
Acquisition
Gateway
Signal Processing
Agent
Quantification
Engine
Clinical Intelligence
Agent
Delivery &
Integration
Report
Literature-based collaborative reasoning
Fully autonomous pipeline — AI agents provide literature-grounded clinical insight

See what 30 years of MRS physics + AI can do.