Challenges
- The client needed to automate the manual coding process to have more consistent and accurate results.
- The client needed to extract data from unstructured and non-standard document formats, including paper, fax, and digital.
- The client also needed to perform a sentiment analysis of the data captured from the patient medical charts.
- Manual review of patient medical charts was a tedious process and prone to errors.
Solutions
- Used Optical Character Recognition to extract data from the scanned medical charts and pulled out relevant medical information.
- Integrated multiple data sources to automate the query process.
- Created an automatic rule-based priority queuing mechanism using NLP.
- Ran the extracted medical data through the NLP engine and captured the sentiment of a particular record by:
- Forming a semantic space word bank for each disease containing a set of important medical keywords that determine the presence of a disease.
- Training the custom models using this word bank.
Tools & Technologies
Python, Open NLP, Leadtools
Key benefits
- Improved process productivity by 40%.
- Increased output rate (15-16 charts per day) using NLP compared to the manual coding rate (10-11 charts per day).
- Provided accuracy rates of 95-98% in identifying medical conditions.
- Reduced overall administration costs.
- Easy access to information expedites care for those in need (patient & doctor).
