Anatomical Model Vector Encoding for LLM Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current technologies lack the ability to effectively interact with complex three-dimensional anatomical models using large language models (LLMs), as these models are typically processed as text and cannot handle the complexity of anatomical structures.

Innovation Solution

An anatomical model is encoded into a data structure with vector representations that can be processed by an LLM algorithm, allowing for manipulation, analysis, and modification of the anatomical model in a hands-free manner.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional text-based processing is used for LLMs, then the LLM can process simple text data, but it cannot handle complex three-dimensional anatomical models

Engineering Contradiction:
Improveability to process anatomical modelsVSAvoidcomplexity of data structure
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent transforms the data representation parameters of anatomical models from traditional text descriptions to vector embeddings. Each anatomical structure is represented as a numerical vector that captures its geometric and topological properties, enabling LLMs to process complex 3D structures through mathematical operations on these vectors rather than attempting to describe them textually.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces an intermediary encoding system that translates complex anatomical model data into a format suitable for LLM processing. This intermediary layer converts geometric and topological information into vector representations that serve as a bridge between the complex anatomical data and the LLM's text-based processing capabilities.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Ease of operation

If hand-operated input devices are used to manipulate anatomical models, then precise control is achieved, but the operator's hand cannot be used in ongoing medical procedures

Engineering Contradiction:
Improvehands-free operation capabilityVSAvoidmanipulation precision
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent replaces the mechanical interaction between hand and input device with a voice-based interaction system. Speech-to-text translation converts verbal commands into digital signals that manipulate the anatomical model, substituting the mechanical control paradigm with an acoustic one that allows hands-free operation while maintaining precision through voice recognition technology.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Loss of information

If a single two-dimensional image is used to represent anatomical structures, then simple image processing is sufficient, but complex anatomical structures cannot be captured

Engineering Contradiction:
Improveanatomical structure informationVSAvoidmodel complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent transitions from two-dimensional image representations to three-dimensional vector-based models. By representing anatomical structures in three-dimensional vector space, the system captures depth, volume, and spatial relationships that are lost in 2D images, while the vector representation maintains computational efficiency for processing.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4571781A1Encoding a model for processing using a large language model
Publication Date: 2025.06.18 KONINKLIJKE PHILIPS NV
  • EP4571781A1 patent drawingFigure 1
  • EP4571781A1 patent drawingFigure 2
  • EP4571781A1 patent drawingFigure 3

AI summary

According to an aspect, there is provided a computer-implemented method (200) for encoding an anatomical model of an anatomical structure of a subject into a data structure suitable for processing using a large language model, LLM, algorithm, the method comprising obtaining (202) an anatomical model of the anatomical structure of the subject, the anatomical model comprising a plurality of model elements; converting (204) a first set of model elements of the plurality of model elements of the anatomical model into a first vector representation capable of being processed using an LLM algorithm; and composing (206) a data structure representative of the plurality of model elements of the anatomical model based on the first vector representation; wherein the data structure comprises an encoded version of the anatomical model of the anatomical structure configured to be processed using the LLM algorithm.