
3 minute read
/// ADVANCING HEALTHCARE WITH NATURAL LANGUAGE PROCESSING
HOW AI IS USED TO IMPROVE MEDICAL PROCESSES
Today, Artificial Intelligence (AI) finds us in many aspects of life. While we use AI daily, such as when asking popular language models (like ChatGPT or Claude) to help us plan a trip, summarize a long text, draft an email at work, find a recipe before cooking dinner, or even help us with financial decisions, there are many uses we are not familiar with, mostly those employed by professionals. One aspect not many are aware of is the ever-growing use of AI in medicine; from analyzing clinical data to diagnostics to treatment, AI is more relevant than ever.
According to Dr. Kfir Bar from the Efi Arazi School of Computer Science, when it comes to Hebrew, language models have two basic limitations. While large language models (LLMs) exist for Hebrew, their performance lags behind models for English and other highly resourced languages. This limitation becomes even more pronounced when dealing with text in specialized domains, such as healthcare, where models must be adapted and trained to understand domain-specific jargon and nuances. Dr. Bar has been collaborating with hospitals and mental health centers across Israel to address the challenge of handling complex patient records. His work focuses on transforming this vast clinical data into concise, structured summaries that are both practical and accessible.
In cooperation with The I-Medata AI Center at the Tel Aviv Sourasky Medical Center (Ichilov), Dr. Bar has been developing a Hebrew language model (similar to ChatGPT) that is being adapted for the medical domain by feeding it with the Hebrew jargon of electronic medical records from the hospital. Three use cases came to Dr. Bar’s mind when he began developing the model: First, it can be used to develop a system that recommends different courses of treatment, tailor-made for each patient, based on their on-file medical history; second, it can assist in creating a “patient journey” by building a timeline with all the major medical events the patient has had based on their medical history and offer possible relationships between these events; and lastly, it can sharpen the medical history to a summary that provides only the most relevant details from the entire file, like past medications and their side effects.
According to Dr. Bar, “With the help of medical language models, I believe it will be possible to improve the processes of diagnosis and treatment. This should happen with the utmost caution as we deal with sensitive medical data while maintaining the patients’ privacy.”
