Overview
Sirma developed Medrec:M, an innovative ambient listening solution designed to automate documentation of doctor-patient interactions in clinical settings. By recording conversations and automatically generating outpatient notes, Medrec:M significantly reduces the administrative burden on physicians while improving the accuracy and efficiency of clinical documentation.
The Challenge
Healthcare providers often face overwhelming documentation demands that detract from the time they spend on patient care. Manual note-taking is time-consuming and prone to errors or omissions, leading to clinician burnout and inconsistent record quality. Existing AI-powered documentation tools tend to be expensive or lack regulatory approval, limiting accessibility and adoption. The challenge was to deliver a cost-effective, AI-approved system capable of ambient listening and real-time, high-fidelity documentation within the regulatory frameworks of the healthcare industry.
The Project Scope
Sirma was tasked with designing and deploying an ambient listening system adapted for live clinical environments to:
- Passively record doctor-patient conversations during outpatient visits without disrupting clinical workflows;
- Apply natural language processing (NLP) and speech-to-text AI algorithms to transcribe and semantically interpret conversations;
- Auto-generate comprehensive outpatient notes structured for medical records and compliance needs;
- Ensure affordable pricing to increase accessibility compared to competing solutions in the U.S. market;
- Achieve regulatory acceptance for AI use cases related to diagnostics and ambient clinical notes.
The Solution
Sirma developed Medrec:M, leveraging advanced ambient listening AI technologies, including:
- Accurate, low-latency speech recognition to capture clinical dialogue in real time;
- Context-aware NLP models trained on medical lexicons to extract relevant clinical facts and context from conversations;
- AI-driven natural language generation to automatically draft structured outpatient notes tailored to physicians’ documentation standards;
- Cloud-based architecture ensuring secure data handling, privacy, and compliance with healthcare regulations;
- User-friendly interfaces for clinicians to review, edit, and finalize notes with minimal effort.
Results
- The system drastically reduced physicians’ administrative workload by automating documentation of patient encounters;
- Medrec:M is priced at approximately $70 per use, a fraction of the $700 cost for comparable US-based AI documentation solutions, widening accessibility and adoption potential;
- It was positioned among only two AI applications approved in the U.S. for clinical diagnostics and ambient note-taking use cases, underscoring its regulatory compliance and innovation leadership;
- Clinicians reported improved satisfaction and efficiency, with more time dedicated to patient care rather than paperwork;
- The solution demonstrated scalability to multiple clinical sites and integration with existing electronic health record systems.
Technologies
The initial outpatient note drafts in Medrec:M are generated using a robust combination of supervised machine learning algorithms and large language models (LLMs) tuned explicitly for clinical language understanding and documentation.
- Advanced Speech-to-Text AI Models: Designed for clinical vocabulary and ambient noise, these models accurately transcribe doctor-patient conversations in real time.
- Natural Language Processing (NLP): Specialized frameworks extract and interpret medical entities and relevant clinical information from unstructured conversation data.
- Natural Language Generation (NLG): AI-driven generation produces coherent and contextually precise outpatient notes automatically, structured to meet documentation standards.
- Secure Cloud Infrastructure: HIPAA-compliant environments ensure data privacy, security, and regulatory compliance throughout processing and storage.
- Integration APIs: Seamless connectivity with hospital electronic medical record (EMR) systems and telehealth platforms enables smooth data exchange and workflow integration.
- Cost-Efficient Architecture: Optimized system design balances high performance and scalability with affordable pricing, expanding accessibility.
- Key supervised AI algorithms powering this solution include:
- Transformer-based Models: Fine-tuned variants of BART, BioBERT, and other bidirectional transformer architectures excel at converting clinical dialogues into accurate, readable notes.
- Sequence-to-Sequence Models: These translate speech transcripts into summarized clinical notes while preserving relevant clinical context and entities.
- Named Entity Recognition (NER) and Relation Extraction (RE): Supervised models trained on clinical data identify symptoms, diagnoses, treatments, and relationships within conversations, enriching note content with structured medical information.
- Section-wise Model Training with Adapters: Modular training for specific clinical note sections (e.g., Subjective, Objective, Assessment, Plan) enhances precision and enables flexible note composition.
- Fine-tuned Large Language Models (LLMs): Specialized models to generate structured K-SOAP-format clinical notes, integrating domain-specific medical knowledge for high-fidelity output.
This AI technology stack enables Medrec:M to automate clinical documentation, alleviating physicians’ administrative burden and enhancing patient care quality while maintaining compliance and cost-effectiveness. It effectively transforms complex, unstructured conversational data into structured draft notes, thereby reducing documentation time and improving the quality of clinical records.
Sirma’s Partnership with the client
Sirma has experience in developing AI-driven ambient clinical documentation solutions that enhance healthcare delivery and significantly reduce administrative burdens. The solution not only adheres to stringent regulatory standards but is also equally suitable for use in hospitals as well as in telehealth settings, making it more accessible and effective worldwide.
