XNews

Mirobody Open-Source Health Record Unifies Lab Data Across Sources

Mirobody Open-Source Health Record Unifies Lab Data Across Sources

Theta Health's open-source Mirobody (Apache-2.0) was tested with virtual data for its ability to recognize the same lab measurement across different reports. The tool converted 193 mg/dL to 4.99 mmol/L locally and mapped both '总胆固醇' and 'total cholesterol' to LOINC code 2093-3 without calling a model or using an API key. After importing a PDF and a CSV report, it plotted a cholesterol trend of 4.60, 4.45 and 4.38 mmol/L, and it declined to assign a code to '血脂' because that term denotes a test category rather than a single reading.

Event version 1.

Key insights

Local unit conversion without a model

Mirobody converted 193 mg/dL to 4.99 mmol/L locally. The step used no model call and no API key.

Cross-language LOINC mapping

The tool mapped '总胆固醇' and 'total cholesterol' to the same international lab code, LOINC 2093-3. It also treated 'Total Cholesterol-TC' and 'Cholesterol, Total' as the same measurement.

Trend built from multiple report formats

After a PDF and a CSV report were imported, a cholesterol query returned a trend of 4.60, 4.45 and 4.38 mmol/L. Each reading can be opened to view its source file.

No code assigned to a test category

When given '血脂 5.0 mmol/L', the tool did not assign a code because '血脂' is a category of tests rather than a single reading.

Public name-recognition cases passed

Version 1.6.0 passed all 347 public name-recognition cases in the repository. This reflects performance on public cases only and does not represent clinical accuracy.

Source posts