AI in Medical Language Processing: Transforming Unstructured Clinical Data

AI Amplified with host Dr. Heather Bassett

Season 1: Episode 9

Guest Tim O’Connell, a radiologist and co-founder of emtelligent, a company specializing in processing unstructured clinical data.

AI in Radiology: Tim’s Journey

Podcast host Dr. Heather Bassett interviewed Dr. Tim O’Connell, a radiologist and co-founder of emtelligent, a company specializing in processing unstructured clinical data. They discussed Tim’s background in enterprise IT before transitioning to medicine and his work in medical informatics at the University of British Columbia. Tim’s experience with AI in radiology, particularly his collaboration with Dr. Anoop Sarkar, led the duo to co-found emtelligent. Their work together resulted in the creation of emtelliPro®, the company’s deep learning-based medical language understanding engine. Tim’s interest in using AI to improve care delivery is rooted in his personal experience as a clinician, and in helping other clinicians better understand patient stories amidst increasing imaging volumes.

AI in Radiology Career Insights

Tim and Heather discussed their career paths, with Tim sharing his transition from engineering to medicine and radiology, while Heather mentioned her background in technology and medicine. They explored the potential of AI in radiology, with Tim expressing excitement about AI’s ability to assist with detection tasks and provide a second set of eyes for diagnostic imaging. Tim also noted that AI models have room for improvement in interpreting data and synthesizing information from multiple sources, an area where human expertise currently excels.

AI in Medical Imaging Evolution

Tim and Heather discussed the evolution of AI in medical imaging and clinical data analysis. Tim shared his background in radiology and imaging informatics, highlighting the challenges of interpreting medical images and the potential of AI to assist in detecting anomalies. Heather emphasized the complexity of clinical documentation and the improvements brought by generative AI and large language models in parsing and contextualizing medical records. Both agreed on the transformative potential of AI in enhancing diagnostic accuracy and clinical workflows.

Deep Learning in Medical NLP

Tim explained the evolution of deep learning and transformer models, highlighting their impact on natural language processing tasks and their current ability to achieve high accuracy levels — exceeding 98-99% for specific tasks like chart review and data extraction. He emphasized emtelligent’s differentiators, including determinism, scalability to process large volumes of documents, cost-effectiveness, and the ability to operate within customers’ clouds for security and confidentiality. Tim also noted that many medical problems are long-tail issues, where the final 5% of accuracy can be challenging to achieve.

AI in Healthcare: Human Context

Tim discussed emtelligent’s software capabilities for processing unstructured clinical data, including handling various data formats and enabling interoperability through ontologies like SNOMED (Systemized Nomenclature of Medicine) Clinical Terminology — the most comprehensive, multilingual, clinical healthcare terminology in the world, used to record clinical information in the patient record. He emphasized the importance of human-in-the-loop AI systems in healthcare to ensure safety and reliability, while Heather highlighted the need to make AI tools valuable to physicians to drive adoption. They agreed that while AI can transform healthcare workflows, human context and judgment remain crucial.