Interpretacija i vizualizacija odluka strojno učenog modela metodom propagacije relevantnosti prilikom otkrivanja autizma pomoću postupka ASD-DiagNet

master's thesis

• Title (english): Interpretation and Visualization of the Decisions of the Machine-learned model Using Relevance Propagation Method When Detecting Autism Using ASD-DiagNet Procedure
• Author: Lara Granoša
• Mentor: Marin Šilić (mentor)
• Committee member: Marin Šilić (predsjednik povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Committee member: Klemo Vladimir (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-04, Croatia

Abstract

U ovom je radu provedena interpretacija i vizualizacija odluka modela ASD-DiagNet za predikciju poremećaja autizma. Analiza je provedena na ABIDE podatkovnom skupu koji je također korišten za treniranje i evaluaciju modela strojnog učenja. Kao metoda interpretacije analizirana je važnost pojedinog ulaza i njegov utjecaj na odluku modela. Za određivanja relevantosti pojedinog ulaza, korištena je metoda propagacije relevantosti po slojevima (LRP). Ova je metoda uspješno primjenjena na model ASD-DiagNet te je pomoću nje dobivena informacija o tome koliko je povezanost dviju regija mozga relevantna za predikciju modela. Pojačana relevantost korelacije dviju regija može biti indikacija promijenjene funkcijske povezanosti te može biti interpretirana kao dokaz za poremećaj autizma. Područja čija je povezanost identeficirana kao relevantna nalaze se u parijetalnom režnju mozga. On je odgovoran za funkcionalnosti koje su narušene kod osoba s autizmom te rezultati na taj način odgovaraju simptomima osoba s ovim poremećajem. Metoda propagacije relevantosti po slojevima može biti primjenjena i na druge modele te na taj način omogućiti dodatno razumijevanje određenih bolesti.

Abstract (english)

In this thesis, interpretation and visualization of decisions of the ASD-DiagNet model for the prediction of autism disorders was carried out. The analysis was performed on the ABIDE dataset, which was also used to train and evaluate the machine learning model. As a method of interpretation, the importance of each input and its influence on the decision of the model was analyzed. To determine the relevance of an individual input, the layer-wise relevance propagation method (LRP) was used. This method was successfully applied to the ASD-DiagNet model, and it was used to obtain information on how relevant the connection of 2 brain regions is for the prediction of the model. Increased relevance of the correlation between two regions may be an indication of altered functional connectivity and may be interpreted as evidence for an autism disorder. Areas whose connectivity has been identified as relevant are located in the parietal lobe of the brain. It is responsible for the functions that are impaired in people with autism, and the results thus correspond to the symptoms of people with this disorder. The method of layer-wise relevance propagation can also be applied to other models and enable additional understanding of certain diseases.

• Keywords: Autizam; ABIDE; interpretacija; objašnjavanje; LRP; vizualizacija
• Keywords (english): Autism disorder; ABIDE; interpretation; explanation; LRP; vizualisation
• Language: croatian
• URN:NBN: urn:nbn:hr:168:180026
• 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: 2023-10-11 08:56:02