Medical Data Extraction With LLM Context Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing medical documentation systems require significant time and effort due to the need for manual conversion of spoken language into written form, and large language models often misinterpret domain-specific medical language due to insufficient contextual understanding.
Innovation Solution
A computer-implemented method using a generative pretrained transformer to process medical data, including audio, video, and image data, and generate automated documentation by preprocessing and tokenizing input data to maintain contextual clarity, with reduction operations to manage token limits, and a multi-agent system for optimizing therapy plans.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual conversion of spoken language into written form is used for medical documentation, then documentation completeness is improved, but time consumption and manual effort increase significantly
Solution Approach 1:
The system enables self-service documentation by automatically generating medical documentation from spoken language inputs without requiring manual conversion. The large language model processes audio inputs, extracts medical data, and produces structured documentation autonomously, eliminating the need for manual transcription while maintaining documentation completeness.
Solution Approach 2:
The patent replaces the mechanical manual transcription process with an automated computational system. Audio inputs are converted to text and processed by a large language model that automatically extracts medical information and generates documentation, substituting human manual effort with an intelligent automated system.
2Productivity
If large language models are used to process medical data, then processing speed is improved, but accuracy decreases due to misinterpretation of domain-specific medical language
Solution Approach 1:
The system changes the parameters of the large language model by fine-tuning it with domain-specific medical data and terminology. This adaptation modifies the model's behavior to better understand and accurately process medical language, thereby maintaining high processing speed while improving accuracy in the medical domain.
Solution Approach 2:
The patent introduces an intermediary processing layer that bridges general large language models and domain-specific medical data. This intermediary layer includes preprocessing steps, domain-specific tokenization, and contextual understanding mechanisms that enable accurate interpretation of medical terminology while maintaining fast processing speeds.
3Quantity of substance
If comprehensive medical data is collected during medical situations, then data completeness is improved, but data processing complexity increases
Solution Approach 1:
The system segments comprehensive medical data into distinct categories and structures during processing. The large language model divides incoming data streams into manageable components such as patient information, diagnosis, treatment details, and observations, making complex data easier to process and store while maintaining completeness.
Solution Approach 2:
The patent applies preliminary action by pre-processing and structuring medical data before main analysis. Data is organized, validated, and categorized in advance using domain-specific rules and templates, reducing the complexity of subsequent processing steps while ensuring all necessary data elements are captured.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method comprises: evaluating a received signal to be indicative of a medical situation of a person; then acquiring input data during the medical situation; preprocessing the input data to obtain preprocessed input data being configured as a subsequent input of a large language model; extracting medical data representing information on the person and at least one of a medical diagnosis of the person, a medical therapy of the person and a medical condition of the person from the preprocessed input data using the large language model in accordance with an extraction task; and generating at least one output based on the extracted medical data using the large language model.