Clinical Scorecard: AI Reconstructs Molecular Structures From Simulated TERS Images

At a Glance

Key Highlights

• SMARTERS converts hyperspectral TERS data into atomic position maps.
• Achieved a mean Dice similarity coefficient of 0.842 on the test set.
• Model predicts atomic positions and elemental identities for hydrogen, carbon, nitrogen, and oxygen.
• Performance declines for non-planar molecules due to weaker signals.
• Limitations noted when applying model to experimental TERS images.

Guideline-Based Recommendations

Diagnosis

Management

Monitoring & Follow-up

Risks

• Substantial gap between simulated and real measurements.
• Differences in experimental conditions affecting model accuracy.

Patient & Prescribing Data

Clinical Best Practices

• Utilize SMARTERS for molecular geometry reconstruction from TERS data.
• Consider limitations of model when applying to experimental data.

Related Resources & Content

• PRX Intelligence Study