Veliki jezikovni modeli na področju medicine in razvoj aplikacije Home Doctor

Zadobovšek, Matic (author), Rožanc, Igor (mentor), Gams, Matjaž (comentor)

Diplomsko delo se osredotoča na preučevanje možnosti vključevanja velikih jezikovnih modelov na področje medicine. V zadnjih letih smo bili priča skokoviti rasti in uporabi le-teh predvsem v obliki klepetalnikov, ki uporabnikom pomagajo pri opravljanju številnih nalog. Raziskave kažejo, da se bo primanjkljaj zdravstvenega kadra v prihodnje le še povečeval, trenutni zdravstveni delavci pa so vse bolj obremenjeni. V okviru diplomskega dela je bila razvita aplikacija, ki se osredotoča na omenjeno problematiko. Glavni motiv je razbremenitev zdravstvenih izvedencev na način, da pacientom omogočimo enostavnejši dostop do ustreznega zdravstvenega svetovanja. Pri tem smo trenutno najzmogljivejši veliki jezikovni model GPT-4o obogatili s specifičnim znanjem s področja slovenskega zdravstva. S tem uporabnikom omogočimo dostop do preverjenih informacij, ki so koristne za slovenskega državljana. Uporabniki komunicirajo s klepetalnikom in pridobivajo odgovore na svoja zdravstvena vprašanja. Hkrati je uporabnikom na voljo dostop do telemedicine, ki celoten pogovor posreduje osebnemu zdravniku. Zdravniku so znotraj aplikacije na voljo številna orodja, ki poskrbijo za hitrejšo obravnavo in samodejno ustvarjanje poročil. Z razvito aplikacijo je bil storjen korak k rešitvi, ki na eni strani pacientom omogoča enostavno pridobitev zdravstvenih informacij, na drugi strani pa zdravnikom z naborom vgrajenih orodij pohitri postopek obravnave pacienta. Testni uporabniki so bili predstavljeni rešitvi naklonjeni. Menili so, da ima aplikacija potencial za nadaljnji razvoj in kasneje tudi uporabo v slovenskem zdravstvu.

• Language: Slovenian
• Keywords: veliki jezikovni modeli, obdelava naravnega jezika, slovensko zdravstvo, arhitektura transformer, RAG, spletna aplikacija, mobilna aplikacija
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-161312
• Publication date in RUL: 09.09.2024

Secondary language

• Language: English
• Title: Large language models in the field of medicine and the development of the Home Doctor application

Abstract

This thesis focuses on exploring the possibilities of integration of large language models in the medical domain. In recent years, we have witnessed an explosion in the growth and use of these, especially in the form of chatbots, which help users to perform a wide range of tasks. Research shows that the shortage of healthcare professionals will only increase in the future and that current healthcare professionals are under increasing pressure. In the framework of the thesis, an application has been developed that focuses on these issues. The main motive is to relieve the burden on medical experts with easier access to appropriate medical advice for patients. In doing so, we have enriched the currently most powerful large language model GPT-4o with specific knowledge in the field of Slovenian healthcare to enable users to easily obtain verified information. Users interact with the chatbot and get answers to their health questions. At the same time, they have access to telemedicine, which relays the entire conversation to a personal doctor. The doctor has a number of tools to speed up treatment and automatically generate reports within the app. The developed application is a step towards a solution that allows patients to easily obtain information on Slovenian healthcare, on the one hand, and speeds up the process of treating the patient with a set of built-in tools for doctors, on the other. Test users were in favour of the solution. They recognized the application's potential for further development and later use in Slovenian healthcare.

• Keywords: large language models, natural language processing, Slovenian healthcare, transformer architecture, RAG, web application, mobile application