Project Overview

MOATiT, an industry leader in providing med-tech solutions to the healthcare industry, sought to leverage AWS AI services for automatic scribing of doctor-patient conversations in a HIPAA-compliant manner. The proposed solution involved a combination of AWS services across compute, storage, and machine learning. Avahi, an advanced tier AWS partner, was engaged to help MOATiT achieve this transformative goal by utilizing AWS HealthScribe, SageMaker, Bedrock, S3, RDS, Lambda, and API Gateway.

Challenges

MOATiT faced several challenges in implementing an automatic scribing solution:

• HIPAA Compliance: Ensuring that the solution complies with strict healthcare data privacy regulations.
• Real-Time Processing: Capturing and transcribing doctor-patient conversations in near real-time.
• Accuracy: Achieving high accuracy in transcriptions, including medical terminology and context.
• Scalability: Building an infrastructure capable of handling large volumes of audio data and user requests.
• Integration: Seamlessly integrating the new solution with existing systems and workflows.

Solution

Avahi proposed a comprehensive solution to address these challenges, focusing on developing a production grade system for an automatic scribing tool. The project was executed in phases, starting with discovery and planning, followed by design and development, validation and QA, and finally, an executive presentation and handoff.

Key Deliverables

• ETL Processes: Performing ETL on raw data to prepare it for model training and API calls.
• Post-Processing Jobs: Running post-processing jobs based on confidence thresholds selected by MOATiT.
• Validation: Ensuring the accuracy, bias mitigation, and scalability of the solution using industry-standard tools.
• Qualitative Review: An analysis report based on manual review to identify common issues and patterns.

Solution Architecture

The solution architecture leveraged various AWS services to implement the deliverables and ensure a robust, scalable platform.

Data Acquisition and Storage

• Amazon S3: Used for storing raw audio files, processed transcriptions, and other related data. S3’s scalability and durability ensured efficient handling of large volumes of data.
• Amazon RDS: Provided a scalable and managed database service for storing structured data, including transcription metadata and medical coding information.

Event-Driven Processing

• AWS Lambda: Deployed to run event-driven functions that process audio data and manage various stages of the pipeline. Lambda functions were used to fetch new audio files, trigger transcription jobs, and update/store outputs.

Machine Learning and AI

• AWS HealthScribe: Employed for scalable managed Generative AI medical scribing services, ensuring efficient and accurate transcriptions.
• AWS SageMaker: Utilized for developing, training, and deploying machine learning models that enhance the transcription accuracy and handle complex medical terminology.
• AWS Bedrock: Used for scalable managed Generative AI services, supporting the overall AI infrastructure.

API Management

• Amazon API Gateway: Provided a managed service to create, publish, maintain, and secure APIs. This facilitated seamless communication between the front-end applications and backend services, allowing easy access to transcribed data.

Monitoring and Logging

• AWS CloudWatch: Implemented for comprehensive monitoring and logging. CloudWatch provided real-time insights into system performance, user interactions, and the health of the deployed models.

Text Analysis

• AWS Comprehend Medical: Used to extract information from unstructured medical text accurately, identifying key medical entities and relationships within the transcriptions

Project Impact

Enhanced Efficiency

The AI-powered scribing tool significantly improved the efficiency of capturing and transcribing doctor-patient conversations, reducing the time and effort required for manual transcription.

Compliance and Security

By leveraging AWS HealthScribe and other AWS services, the solution ensured HIPAA compliance, protecting sensitive healthcare data while maintaining high levels of security.

Scalability and Flexibility

The architecture designed by Avahi ensured that MOATiT’s platform was scalable, capable of handling increased data volumes and user demands. This provided flexibility for future enhancements and growth.

Improved Accuracy

The use of advanced machine learning models in AWS SageMaker and AWS Comprehend Medical enhanced the accuracy of transcriptions, including the detection of medical coding such as ICD-10 codes.

Detailed Insights and Analytics

The deployment of AWS CloudWatch and other monitoring tools provided MOATiT with detailed insights into system performance, transcription accuracy, and user behavior. These analytics were crucial for continuous improvement and optimization of the platform.

Conclusion

The collaboration between MOATiT and Avahi led to the successful development of an AI-powered medical scribing tool. Leveraging AWS services, MOATiT enhanced its transcription process, ensuring efficiency, accuracy, and compliance with healthcare regulations. Avahi’s expertise in AWS solutions played a pivotal role in this transformation, demonstrating the power of Generative AI and machine learning in advancing med-tech solutions.