LLM AI Agent for Tumor Board Discussion Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current machine learning language models (LLMs) lack self-awareness of their accuracy and tend to 'hallucinate' content, making them impractical for high-stakes applications like Tumor Board meetings where precision is critical for developing effective radiotherapy treatment plans.

Innovation Solution

Implementing a collaborative Tumor Board Application (TBA) that utilizes an LLM-powered AI agent to interact with Tumor Board discussions, collecting and analyzing data to improve information capture and analysis, and presenting accurate and relevant information to enhance treatment planning decisions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If current LLMs are used to assist Tumor Board discussions, then information capture and analysis efficiency is improved, but accuracy and reliability deteriorate due to hallucination and lack of self-awareness

Engineering Contradiction:

Improveinformation capture and analysis efficiencyVSAvoidaccuracy of generated information

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent introduces an intermediary verification layer between the LLM and the Tumor Board discussion. This intermediary consists of multiple validation mechanisms including fact-checking modules, cross-referencing with medical databases, and confidence threshold filtering that prevent hallucinated content from being presented to medical professionals, thus maintaining reliability while preserving productivity benefits

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback loops where the LLM's outputs are continuously evaluated against ground truth medical data and previous Tumor Board decisions. Accuracy metrics are tracked and used to refine the model's responses in real-time, creating a self-correcting system that improves reliability without sacrificing the efficiency gains from automated information processing

Inventive Principle:

Principle #23Feedback

2Productivity

If LLMs are deployed for treatment planning assistance, then discussion efficiency is improved, but precision deteriorates due to ambiguous responses and hallucinated content

Engineering Contradiction:

Improvediscussion efficiencyVSAvoidprecision of treatment recommendations

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent applies partial action by having the LLM generate only high-confidence responses that meet predetermined precision thresholds. Low-confidence or ambiguous responses are flagged for human review rather than being presented as definitive recommendations, ensuring that only precise information contributes to treatment planning while maintaining efficient discussion flow

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system performs preliminary verification and validation of LLM-generated content before presenting it to the Tumor Board. This includes pre-checking against established medical guidelines, cross-referencing with patient-specific data, and filtering out potentially hallucinated content in advance, thereby ensuring precision is maintained while efficiency is improved

Inventive Principle:

Principle #10Preliminary action

3Manufacturing precision

If multiple rounds of interaction with computer models are performed to verify treatment plans, then treatment plan quality is improved, but time consumption increases

Engineering Contradiction:

Improvetreatment plan qualityVSAvoidtime for multiple verification rounds

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent implements preliminary verification mechanisms that validate treatment plan components before they are finalized. The LLM generates multiple candidate plans with confidence scores, and the highest-confidence plans that meet quality thresholds are selected immediately, eliminating the need for multiple iterative verification rounds and reducing time loss while maintaining high treatment plan quality

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables self-service verification where the LLM automatically validates its own generated treatment plans against stored medical protocols, patient data, and quality criteria. This self-validation process occurs in real-time during plan generation, providing immediate quality assurance without requiring separate verification rounds, thus improving treatment plan quality while minimizing time consumption

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20250200331A1Using language models to assist tumor board discussion

Publication Date: 2025.06.19 SIEMENS HEALTHINEERS INTERNATIONAL AG

AI summary

Embodiments described herein provide for implementing a language model in a use-case for Tumor Board meetings for radiotherapy treatment planning (RTTP) efforts and treatment planning assistance, which requires high-levels of accuracy and/or precision in the outputs generated by the LLM and presented to members of the Tumor Board participating in a Tumor Board discussion, which may include live meetings or asynchronous online discussions. Tumor Board Application (TBA) software collects from discussions of a Tumor Board meeting to train the LLM on predicting outputs that contribute information about the patient, proposed RTTP, or aspects of the patient treatment. An AI agent participates in the Tumor Board discussion to ingest the inputs of the members of the Tumor Board and output the responsive text produced by the LLM, thereby allowing the LLM-powered AI-agent to interact with Tumor Board discussions.