VoxRad: Building an open-source locally-hosted radiology reporting system.

Abstract

VoxRad is an open-source application designed to enhance radiology reporting by leveraging generative AI. Utilizing locally hosted Automatic Speech Recognition (ASR) and Large Language Models (LLM), VoxRad enables continuous dictation, transcribing reports into standardized formats with high accuracy, efficiency, and data security. The modular design allows flexible integration of user-selected ASR and LLM models via OpenAI-compatible APIs, ensuring HIPAA compliance with secure local storage of data. Customizable template guided prompting using Chain-of-Thought like systematic processing, and specialized dictionaries further optimize report generation. VoxRad's future aims include healthcare system integration and community-driven template libraries, enhancing its utility for the medical community.