AI Question Answering System Sub-document Segmentation for PHI Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training AI systems for healthcare or medical QA is challenging due to large knowledge bases and the need to keep patient records segregated to avoid Protected Health Information (PHI) violations, making it difficult to implement effective QA systems.
Innovation Solution
A QA system using an AI retriever engine to divide the knowledge corpus into sub-documents, encode queries and documents, and identify matching sub-documents for inference, while updating models based on loss functions, reducing the need for access to the entire knowledge base and preventing PHI mixing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the entire knowledge corpus is used for training, then the AI system has access to comprehensive medical information, but patient privacy is compromised and PHI mixing occurs
Solution Approach 1:
The patent divides the knowledge corpus into multiple separate patient record documents, each processed independently. This segmentation allows the AI system to access comprehensive medical information through systematic processing of individual records without mixing or compromising patient privacy, as each document is handled in isolation with proper privacy safeguards.
2Object-affected harmful factors
If the knowledge corpus is divided into sub-documents, then patient privacy is protected, but the AI system loses access to comprehensive information
Solution Approach 1:
The patent implements continuous processing of segmented patient records through systematic retrieval, reading, and aggregation of information across multiple sub-documents. This continuity ensures that while each document is processed privately and separately, the AI system accumulates comprehensive medical information through the continuous workflow of processing the entire corpus in a privacy-preserving manner.
3Measurement precision
If AI models are trained on large knowledge bases, then answer accuracy improves, but training complexity and computational resources increase
Solution Approach 1:
The patent segments the large knowledge corpus into manageable patient record documents and processes them through a modular retrieval and reading pipeline. This segmentation reduces training complexity by breaking down the overwhelming task of processing entire large-scale electronic health record systems into smaller, more manageable units that can be processed sequentially or in batches, while still achieving comprehensive information coverage.
Solution Approach 2:
The patent introduces intermediary components including a document retrieval system and a reading system that act as mediators between the knowledge corpus and the AI model training process. These intermediaries manage the complexity of handling large knowledge bases by implementing structured workflows for document selection, extraction, and processing, thereby simplifying the overall training pipeline while maintaining answer accuracy.
4Quantity of substance
If document-level labels are used, then training data availability increases, but the ability to locate specific information within documents decreases
Solution Approach 1:
The patent extracts specific information and relevant passages from within patient record documents while preserving document-level labels. The reading system identifies and extracts pertinent information segments from each document, making the training data more useful by providing both the label context and the specific supporting information, thereby maintaining information location precision while utilizing document-level labeling.
Data Source
AI summary
A method includes discarding a current knowledge corpus; selecting a new knowledge corpus; performing operations as follows using an Artificial Intelligence (Al) retriever engine: dividing the new knowledge corpus into a plurality of sub-documents; encoding a query for the plurality of sub-documents using a query encoding model; encoding each of the plurality of sub-documents using a document encoding model; and determining at least one matching sub-document of the plurality of sub-documents that is a match for containing an answer to the query based on the encoded query and each of the plurality of encoded sub-documents; performing operations as follows using an AI reader engine: generating an inference about the answer to the query based on a concatenation of each of the at least one matching sub-document with the query, each of the at least one matching sub-document having an associated reader loss function result for the inference; identifying one of the at least one matching sub-document having a lowest reader loss function result; and associating the identified one of the at least one matching sub-document with a truth label for the query.