Brain Image Semantic Vector Reconstruction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for decoding mental content from functional brain images are limited, particularly in generating natural language representations, as they require extensive databases for stimuli and struggle with complex mental states like language text, and are not feasible for reconstructing complex constructs such as clouds of words sized by their probability in the subject's mind.

Innovation Solution

A system and method that uses a combination of linguistic semantic vector representation, basis learning, and text generation modules to decode mental content from functional brain images, where semantic vectors are assigned to training data, mapped to brain activation patterns, and used to generate natural language sentences, enabling the reconstruction of mental content from unknown stimuli without relying on specific input types or databases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If generative brain decoding is used to reconstruct mental content, then the ability to represent complex mental states is improved, but the requirement for extensive databases with most conceivable stimuli increases

Engineering Contradiction:
Improveability to represent complex mental statesVSAvoiddatabase size
Core Design Contradiction:
Adaptability or versatilityVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential semantic features from stimuli rather than storing complete stimulus databases. By mapping brain activity patterns to semantic vectors that capture the core meaning, the system eliminates the need for extensive stimulus databases while maintaining the ability to represent complex mental states.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces semantic vectors as an intermediary representation between brain activity patterns and mental content. Instead of directly mapping to specific stimuli from a database, the system uses semantic vectors to bridge the gap, allowing generalization to unseen stimuli while requiring minimal training data.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If a database with most conceivable stimuli is used for generative brain decoding, then the coverage of mental content is improved, but the feasibility and complexity of the system decreases

Engineering Contradiction:
Improvecoverage of mental contentVSAvoidsystem feasibility
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system extracts only the essential semantic dimensions needed to represent mental content, rather than maintaining comprehensive stimulus databases. This extraction approach maintains coverage of mental content while dramatically reducing system complexity and improving feasibility.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter space from discrete stimulus categories to continuous semantic vectors. This transformation allows the system to represent any mental content within the semantic space without requiring explicit database entries, thereby reducing complexity while maintaining versatility.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If language text is decoded using traditional methods, then the accuracy for known stimuli is improved, but the applicability to unknown or complex mental constructs decreases

Engineering Contradiction:
Improvedecoding accuracyVSAvoidapplicability to unknown stimuli
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent uses semantic vectors as an intermediary that enables generalization from known to unknown stimuli. The semantic representation captures the essential meaning that can be applied to any stimulus within the semantic space, maintaining accuracy for known stimuli while extending applicability to unknown or complex mental constructs.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The semantic vector representation serves multiple functions simultaneously: it captures the meaning of known stimuli for accurate decoding, represents unseen stimuli through semantic similarity, and enables generation of novel mental content representations. This multi-functionality resolves the contradiction between precision and versatility.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3364868B1Generating natural language representations of mental content from functional brain images
Publication Date: 2021.05.19 SIEMENS MEDICAL SOLUTIONS USA INC
  • EP3364868B1 patent drawingFigure 1~3
  • EP3364868B1 patent drawingFigure 4~5A
  • EP3364868B1 patent drawingFigure 5B~6

AI summary

By way of introduction, the present embodiments described below include apparatuses and methods for generating natural language representations of mental content from functional brain images. Given functional imaging data acquired while a subject reads a text passage, a reconstruction of the text passage is produced. Linguistic semantic vector representations are assigned (1301) to words, phrases or sentences to be used as training stimuli. Basis learning is performed (1305), using brain imaging data acquired (1303) when a subject is exposed to the training stimuli and the corresponding semantic vectors for training stimuli, to learn an image basis directly. Semantic vector decoding (1309) is performed with functional brain imaging data for test stimuli and using the image basis to generate a semantic vector representing the test imaging stimuli. Text generation (1311) is then performed using the decoded semantic vector representing the test imaging stimuli.