With the recent explosion in the Generative AI domain, like many other industries, healthcare organizations are also exploring ways to integrate LLMs into their workflows. Before we take this discussion any further, I wanted to highlight some healthcare-related LLM applications that already exist:

Med-PaLM (Google AI): This LLM has been trained on a medical text and code dataset. It can perform various medical tasks, in addition to answering questions. Examples are interpreting images, generating radiology reports, and calling genomic variants. Med-PaLM2 is among the models that power MedLM, a family of foundational models fine-tuned for healthcare.

Flan-PaLM (Google AI): This LLM has been trained on code and medical data. Flan-PaLM has attained high performance on various medical benchmarks. Examples of these benchmarks include the Medical Language Understanding Evaluation (MLUE) and the MultiMedBench.

BioGPT-JSL (John Snow Labs): This LLM has been trained on a massive dataset of biomedical literature, scientific papers, and clinical data. It has demonstrated high proficiency in tasks like generating clinical reports, summarizing medical literature, and answering medical questions.

ClinicalBERT (Google AI): This LLM has been trained on clinical text, including electronic health records (EHRs), clinical notes, and medical reports. It can perform tasks like identifying medical entities, classifying medical text, and generating medical reports.

PubMedBERT (NLM): This LLM has been trained on a massive dataset of biomedical literature from PubMed. It can perform tasks like answering questions, summarizing texts, and extracting relations.

The above list may not be exhaustive, but it is a good representation of the functionalities LLM models in the area currently have. These LLMs still have a long journey to cover until they consistently match the expertise of skilled healthcare professionals (consistency is the key).

Yet, there is no doubt that there is a huge potential for integrating LLMs as the third entity in the doctor-patient relationship. In my opinion, these models have the power to re-invent almost all the critical clinical and administrative processes that currently need humans to create original work. Processes include medical coding, diagnosis, patient education and intake, planning patient treatment, medication management, etc.

While the world is much more familiar with Generative AI products like Bard and ChatGPT, I think the powerful aspect of the success of these tools has been the fact that they have shown us the power of the consumer-facing capabilities of LLMs. But the fact is that the LLM landscape, and hence the potential of LLM models is vast. These two popular models are excellent examples of the capabilities of LLMs, but they are a small representation of the potential of LLMs.

In this article series, we will explore the possibilities of leveraging Generative AI in healthcare. This exploration will avoid the technical details but any possibility shared in this article series will be fully within the realm of current LLM and other technological capabilities. The second part of this series will be published on 06/20.