Levo

Levo runs on a Windows radiology workstation. A macro pad and a mouse send key codes, and Levo turns each code into the right command for the image viewer under the pointer or for the PowerScribe report. It knows two PACS viewers and sends each one its own shortcut, so one set of keys works in both.

Radiologists who read at a PACS workstation and dictate into PowerScribe. The workstation build is private and in daily use. A public copy of the earlier RadMode script, with a setup guide, is on rads.icu. Levo was named RadMode until v25.97.

AutoHotkey v2 and a Planck macro pad.

Levo stores no window titles. Report text leaves the workstation only de-identified, and only for the Proofreader check below.

Pre-sign proofreading for a dictated radiology report. Proofreader reads the report before it is signed and flags the errors that speech recognition puts into it. It plugs into Levo as a module, but it is its own project.

A misheard word or a sound-alike, a side or a sex that conflicts with the exam title, a measurement that conflicts with itself, a wrong unit, a negation flipped against another section, a repeated or dropped word, and findings that disagree with the impression. Anything that needs outside knowledge, such as a guideline or a follow-up interval, is out of scope.

These errors are easy to miss in a report you have just dictated. Each issue comes back with the exact quote, the smallest fix and the reason, and one key applies a fix. The check never blocks signing: the radiologist decides. Report text is de-identified on the workstation before it leaves.