Unlocking Clinical Insights with Generative AI

Different tumor types may present in multiple clinically and biologically distinct forms, making standardized data capture challenging 

Differences in pathology report formats and structure across distinct hospital systems adds an additional layer of complexity to standardized data capture

Traditional NLP methods do very well in specific applications, where terms are explicitly defined and nuances in language, grammar, and formatting don't need to be considered

GenAI is robust to nuanced language, grammar, and formatting, built on large language models that enable execution of prompt instructions with interpretation of complex concepts

In this experiment we extracted thymoma tumor size (with units), diagnosis type, TNM staging, and Masaoka / Masaoka-Koga staging

Yes, we created a generalized set of prompt instructions that was robust in extracting accurate insights across pathology reports from 6 distinct hospital systems. Three of these hospital systems were used for defining / refining the prompt, while the remaining 3 were used to validate the generalizability of the prompt. Because we employed thorough validation testing, we expect the generalized prompt to do well in extracting data from pathology report formats not previously tested.

The prompt instructions can be further refined to identify and extract clinical entities in addition to the ones specified in our experiment

To demo or trial this solution, contact your Axtria Clinical Solutions team at axtria_clinicalsolutions@axtria.com