Segmentacija slika magnetske rezonancije i dijagnostika pomoću virtualnog medicinskog asistenta

master's thesis

• Title (english): Magnetic resonance images segmentation and diagnostics using a virtual medical assistant
• Author: Mia Jurdana
• Mentor: Jelena Božek (mentor)
• Committee member: Jelena Božek (predsjednik povjerenstva)
• Committee member: Sonja Grgić (član povjerenstva)
• Committee member: Dario Bojanjac (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-09-13, Croatia

Abstract

U ovom radu razvijen je virtualni medicinski asistent za automatiziranu segmentaciju organa i promjena u tkivu iz slika magnetske rezonancije. Korisnici mogu jednostavno komunicirati s asistentom putem intuitivnog chat sučelja. Istražuje se primjena pretrenirane samo-konfigurirajuće U-net mreže za segmentaciju na raznolikim podacima koji variraju u svojim značajkama poput broja voksela, razmaka među vokselima, broja primjera u skupu za treniranje i broja klasa. Dodatno, asistent koristi napredne tehnike pretraživanja vanjske baze znanstvenih članaka za odgovaranje na postavljena pitanja, čime se dodatno unapređuje preciznost i korisnost njegovih odgovora i smanjuje mogućnost pojave halucinacija. Fokus rada je na integraciji postojećih tehnologija umjetne inteligencije kako bi se stvorio učinkovit sustav za medicinsku dijagnostiku.

Abstract (english)

This work presents the development of a virtual medical assistant designed for the automated segmentation of organs and tissue changes from magnetic resonance scans. The assistant allows users to interact through an intuitive chat interface, enabling seamless communication and access to advanced diagnostic tools. It utilizes a pretrained selfconfiguring U-net model for segmentation, capable of adapting to diverse datasets with varying characteristics such as voxel count, voxel spacing, training set size, and the number of classes. Additionally, the assistant uses advanced retrieval-augmented generation techniques to search external scientific literature databases, providing precise and reliable responses to medical and diagnostic inquiries, while minimizing the risk of misinformation or hallucinations. The focus of the work is on integrating existing artificial intelligence technologies to create an effective system for medical diagnostics.

• Keywords: U-net mreža za segmentaciju; MR segmentacija; virtualni medicinski asistent; analiza medicinskih slika; RAG
• Keywords (english): U-net segmentation network; MRI segmentation; virtual medical assistant; medical image analysis; RAG
• Language: croatian
• URN:NBN: urn:nbn:hr:168:889821
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 20:59:01