Inspiration
The inspiration behind LiveDoc came from our deep-rooted commitment to enhancing healthcare services. We observed the pressing need for a more efficient and accurate Electronic Health Records (EHR) management system in the healthcare industry. Inconsistent data structures and challenging data retrieval processes were impeding the quality of patient care, which motivated us to develop a solution that could revolutionize EHR management.
What it does
LiveDoc is a comprehensive healthcare solution that combines a standardized Electronic Health Records (EHR) model with a cutting-edge domain-centric AI-based search engine. Our platform streamlines EHR management, ensuring data consistency and accuracy. The AI search engine empowers healthcare professionals to access critical patient information swiftly and accurately, ultimately improving diagnosis, treatment, and overall patient care.
How we built it
We created a procedure to standardize EHR data schema using various Clojure data as code paradigms and feed the standardized data to a transformer-based natural language model to perform a smart search on EHR data.
To make the data standardized, we transformed some arbitrary data into domain-centric sub-sections. So for example, we took names, patient information, patient medical issues, and doctor notes and divided them into sub-sections for data consistency.
By making the data consistent, we solved two problems. The first problem is data interoperability, and the second problem is effective search on the EHR data.
Data interoperability will now allow hospitals to use a platform-agnostic EHR data schema, and thus revolutionize the existing EHR systems that are organization-centric. This will avoid data inconsistency results, and allow hospitals to effectively communicate with each other regarding the patient's contextual data. A standardized data schema will also allow the IT departments in the hospitals to create software for medical purposes with a disciplined approach.
Currently, searching EHR data is a problem as it follows a monolithic pattern. Hospital staff has to dive deep into the nested data structure to perform complex searches. Since the data is now divided into domain-centric subsections, it makes transformer models perform question and answering more effectively. As a result, a quick and accurate search of the EHR data can be achieved.
- Clojure allowed us to use the "data as code" paradigm to generate and create a consistent data schema for the EHR data.
Langchain allowed us to train a custom transformer-based model to develop a search engine that is based on the question-and-answer paradigm on the EHR DATA.
We used Flask, and simple HTML/CSS based UI to host our back-end APIs and user interfaces.
Challenges we ran into
Creating a standardized data schema for EHR is a challenge as there are multiple nested medical related fields. We have to carefully analyze EHR data samples and make a generic assumption on which data attributes can be put into a generic EHR data schema.
Accomplishments that we're proud of
- Standardize data format with Clojure. (Not 100% perfect).
- An accurate and efficient search engine for the EHR model.
- A fully functioning web application that hosts a AI-based search engine for the EHR data.
What we learned
Throughout the development of LiveDoc, we learned the critical importance of collaboration between technical experts and healthcare practitioners. Understanding the intricate needs of healthcare professionals and the ever-evolving nature of medical data proved to be instrumental. Additionally, we gained valuable insights into data privacy regulations and the complexities of AI integration in healthcare.
What's next for LiveDoc
Automated data analytics for the EHR data that is not standardized. Try to optimize our search engine for large amount of EHR data (make it more scalable)
Log in or sign up for Devpost to join the conversation.