Automated Medical Transcription
- The application automates the entire medical transcription process.
- Doctors are allowed to dictate notes about a patient encounter using a mobile application.
- The dictated notes get transcribed to text using speech to text solutions that utilize machine learning.
- Transcribed text gets submitted as a task to a queue where an authorized transcriptionist can pick up the task and modify the textual content if any.
- Automated coding and annotation are performed on the corrected content. Codes such as ICD-10 get applied and added to the EMR against the respective fields.
- Relevant information is extracted from the text and made available in a structured fashion to the EMR.
- Once the transcriptionist submits the corrected case, the changes sync back to the EMR.
- The dictation module is a native iOS application.
- Integration with OpenEMR, OpenMRS, Bahmni through SDKs and APIs.
- Speech to text was done using customized algorithms plugged into the Kaldi engine. This was important to enable good speech to text for medical terminologies.
- A Queuing mechanism was set up using a relational database for persistence.
- Workforce module was developed using Node.JS
- User management was done using Passport.JS and has support for LDAP.
- ML Components to extract vitals was primarily done using a custom word embeddings model that consumed dependency trees. Also, external ontologies such as SNOMED-CT, RXNorm etc were integrated into this algorithm and this was used to annotate and enrich the text.
- Java 8
- StanfordNLP Universal Dependencies
- Node.JS with Express