AI Interpreter

The AI Interpreter is the core feature of Gait Sharing, providing the ability to communicate with LLMs (currently ChatGPT) and request automated descriptions of measured gait data.

File Selection

First, the user selects which files to transfer to the LLM:

features.txt: The extracted features informing deviations from normal
Anonymized PDF: Additional patient information not available in the features, such as spatiotemporal parameters

The user will also be asked to provide a description of the uploaded files.

Prompt Engineering

The prompts sent to ChatGPT can be engineered within the tool. There are five different prompt categories available, and prompts can be customized and saved for future use.

Diagnosis Context

The diagnosis context (e.g., cerebral palsy) should be entered in the designated box to guide the LLM in generating a description of the patient’s gait. The output report will be named after the patient; however, this information remains local and is not transmitted to OpenAI.

API Key

To access OpenAI, the user needs a personal API key. Instructions for obtaining this key are provided in a pop-up window within the application.

Note

In case of difficulty acquiring the key, please contact the author for support.

Privacy and Legal Considerations

As mentioned in the application, the responsibility for legal compliance rests with the user. However, uploading anonymized features — not curves, which could serve as biomarkers — along with a user-provided summarized diagnosis to guide the LLM, and receiving a detailed description of the measurements and general cause-and-effect chain, represents the safest approach through this platform.