Mendel, a leader in clinical AI, has announced advancements in its neuro-Symbolic AI system, which automates the identification of patient cohorts from medical records and surpasses GPT -4 in key benchmarks.
Mendel’s unique AI approach combines large language models (LLMs) with its hypergraph reasoning engine. This research demonstrates how the system empowers Automatic Cohort Retrieval (ACR), which is crucial for clinical research and patient care.
Traditional methods of cohort identification are time-consuming and often inaccurate. Mendel’s AI utilizes an LLM trained on the medical text and a reasoning engine informed by medical knowledge, mimicking a clinician’s cognitive process.
Enhanced Performance over Existing Techniques
The research introduces two new reasoning types:
• Structured reasoning: Identifying patients who meet specific criteria.
• Temporal reasoning: Identifying patients over a specified time period.
In benchmark evaluations, Mendel’s Neuro-Symbolic System, Hypercube, outperformed LLM-based systems. For example, Hypercube achieved an F1 score of 62.9 in one benchmark, significantly higher than GPT-4’s 20.8.
This breakthrough has significant implications for clinical research and patient outcomes:
• Faster and more accurate identification of patient cohorts for clinical trials.
• Improved patient stratification for personalized treatment.
• Enhanced efficiency in retrospective studies.
Mendel’s research underscores its commitment to advancing AI in healthcare. This Neuro-Symbolic AI system represents a transformative step towards more robust and scalable clinical reasoning, improving patient care and the efficiency of clinical research.
Jon Napitupulu is Director of Media Relations at The Clinical Trial Vanguard. Jon, a computer data scientist, focuses on the latest clinical trial industry news and trends.