Mendel, a leader in Clinical AI, has announced groundbreaking results from its latest research on Neuro-Symbolic AI.

Their Clinical AI system now excels at automating the identification of patient cohorts from both structured and unstructured EMRs, surpassing benchmarks set by GPT-4 in several key areas.

This achievement is attributed to Mendel’s innovative approach, which integrates large language models (LLMs) with a proprietary hypergraph reasoning engine.

The research showcases Mendel’s ability to significantly advance Automatic Cohort Retrieval (ACR), a critical task in clinical research and patient care.

Unlike traditional methods that rely on manual curation and structured data queries, Mendel’s AI applies sophisticated clinical reasoning to improve the quality and efficiency of cohort identification.

This approach has demonstrated superior performance over existing Retrieval-Augmented Generation (RAG) and LLM techniques.

The milestone this research represents for AI in healthcare: "Our neuro-symbolic architecture not only enhances the effectiveness of patient cohort retrieval but also sets a new standard in clinical reasoning.

By integrating LLMs with hypergraph reasoning, we aim to transform clinical research and enhance patient outcomes."

Key findings include the introduction of two innovative reasoning techniques:

1. Longitudinal Reasoning: Mendel’s neuro-symbolic architecture efficiently handles the longitudinal nature of unstructured EMRs, processing patient records in a single offline pass to construct a comprehensive patient journey. This approach contrasts with LLM-only methods by minimizing repeated querying costs.

2. Large-Scale Reasoning: Integrating real-time hypergraph reasoning with clinical LLMs improves Precision and Recall in cohort retrieval tasks. Unlike LLM-only solutions that scale with database size, Mendel’s approach maintains consistent query costs, making it feasible for healthcare applications.

Mendel’s research introduces a new benchmark for ACR, comparing the performance of RAG, LLM-based solutions, and their neuro-symbolic systems.

The evaluation underscores the transformative potential of their approach, enhancing the accuracy and efficiency of cohort retrieval for precise patient stratification and targeted interventions.

 

 

 

 

 

Source: businesswire.com

Report Abuse