Our technology

Local AI models, built from your data

Our local model-building expertise means every product is tuned to your own data, delivering the best solution for each individual site rather than a one-size-fits-all system.

The model-building process

Validated on your cases before go-live

We will validate a locally-tuned model from your data. 80% of data is used for tuning, 20% set aside as a clean validation test set.

80%Tuning set: your historical cases
20%Held-out validation
Validation on historical data
01Top-1 accuracy

Replay your historical cases and measure how often the top prediction matches the final billed CPT code.

02Automation is clear

Find the confidence threshold where accuracy is 95%+; cases above automate, rest route to review.

03You review & approve

A full per-CPT performance report to validate results and sign off before go-live.

Example results

Top-1 accuracy from a prior site’s historical-case replay

89%Surgical CPT
99%Anesthesia CPT

During validation, your historical coding is held up as the source of truth: every prediction is scored against the code your team billed. Where the model and the billed code disagree, the review often identifies opportunities to improve coding accuracy as well.

Our patent

Automated system and method for assigning billing codes to medical procedures

Our patented process creates local, site-specific AI models trained on historical data, combining natural language processing, machine learning, and large language models to turn clinical text into billable CPT codes.

Exclusively licensed from the University of Michigan.

View the patent

Patent no.
US 11,288,445 B2
Granted
March 29, 2022
Filed
December 10, 2019
Assignee
University of Michigan
License
Exclusive to Decimal Code
Inventors
Michael L. Burns, John Vandervest, Sachin Kheterpal, Nirav Shah, Leif Saager, Jay Jeong, Anik Sinha

Research

Research papers

2020
Classification of Current Procedural Terminology Codes from Electronic Health Record Data Using Machine Learning

Classifies anesthesiology CPT codes from EHR data with up to 96.8% accuracy.

Burns ML, Mathis MR, Vandervest J, et al. · Anesthesiology 132(4):738–749
Read paper
2021
Neural Machine Translation–Based Automated Current Procedural Terminology Classification System Using Procedure Text

Translates surgical procedure text into CPT codes, with top-three accuracy above 95%.

Joo H, Burns M, Kalidaikurichi Lakshmanan SS, Hu Y, Vydiswaran VGV · JMIR Formative Research 5(5):e22461
Read paper

Let’s chat

See your own validation results

We will validate a locally-tuned model from your data. 80% of data is used for tuning, 20% set aside as a clean validation test set.