FosterHealth AI’s HIPAA compliant AI-powered scribe generates fact-checked clinical notes based on conversations between patients and physicians.
Challenges in transcribing multilingual healthcare conversations
Our current customers include leading research institutions, individual practices and small clinics spanning across the United States and India. The conversations between patients and physicians may occur in English, other languages (examples: Hindi, Spanish) or switch between English and other languages. Additionally, the speech processing technology must handle complexities such as crowded outpatient departments with high ambient noise.
Current state-of-the-art transformer architecture based models have performance limitations in scenarios when the conversations occur in languages other than English or switch between English and other languages. Sometimes, they get stuck in the same loop (repeat the same sentence multiple times), miss chunks of data while translating, make spelling mistakes while transcribing information that is medically relevant (example : “Amyatinib” instead of “Imatinib” — while the AI transcribed word is phonetically similar, it constitutes an error). In some instances, they hallucinate inaccurate information that is not presented in the conversational transcript and do not always present information in the custom note format we need. We need technological and interface innovations to address these reliability related issues.
Our design philosophy
Every healthcare operator must find our application trustworthy. They should feel comfortable and confident about our AI model’s outputs every time they use our application to get help with documentation related tasks. To achieve this design goal, we incorporate three main design principles into our product development process:
Design a fallback system that enhances the reliability by actively monitoring the primary AI system’s output
If needed, the AI model should ask the user for help before generating the final output
The user should be able to review the AI outputs in a seamless manner
In this blog, we describe how our application leverages these principles and reliably transcribes ambient multilingual healthcare related conversations.
Speech processing system
We enforce multiple engineering controls to improve the reliability of primary AI. We have three different complex, transformer architecture based AI systems in our primary AI handler. During the decoding stage, we have a fault detection system — if it detects any faults, it calls the second AI system. If the fault detection system detects faults with the second system’s output, we split the audio into smaller chunks and call the third AI system. This approach of using redundant systems to improve reliability is extensively used in other industries which are safety-critical (example: air crafts, automobiles). Additionally, we have an independent fallback system that uses classical natural language processing based algorithms to reliably transcribe medical terms (more information on this can be found in our Medi-Spell checking system blog).
For our enterprise customers, we assist in setting up the right audio infrastructure that can be affixed to the consultation room workstations. This fine-tuning process enhances the fidelity of audio signals the system captures, ultimately improving the reliability of the speech processing system and the overall user experience.
Our goal is to deliver state-of-the-art technology in a reliable and trustworthy manner. We are constantly talking to our users, collaborating with leading research institutes and healthcare experts and continually improving our service. If you have any additional questions or if you want to partner with us, please contact us here.
Comentarios