Aju aktiivsuse analüüs otsuste tegemise vältel
Files
Date
2019
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Orbitofrontaalne ajukoor (OFC) on aju ees istuv piirkond, mille toimimist ei ole suudetud täielikult mõista. Siiski on see selgelt seotud otsuste tegemisega, nagu on näidatud paljudes viimastel aastakümnetel läbi viidud neuroloogiauuringutes. Saez jt [1] on oma viimases uuringus leidnud tõendeid selle kohta, et OFC kõrge sagedusega aktiivsus (HFA) 70-200 Hz vahel on otseselt seotud käitumisreaktsioonidega otsuste tegemisel. Näiteks näitasid Saez jt, et mõned HFA modulatsioonid korreleeruvad inimese valikuga ja tagajärgedega lihtsa kihlveo mängus. Saez jt viisid läbi analüüsi ühe muutujaga lineaarse regressiooniga, ennustades HFA väärtusi korraga ühest ülesandega seotud parameetrist, et leida elektroode, mis kodeerivad otsuste tegemisega seotud informatsiooni. Antud magistritöö keskendus Saez jt tulemuste ja analüüsi laiendamisele, rakendades mitmemõõtmelisi meetodeid, et avastada keerulisi signaale ja olulisi mustreid neuroloogilistes andmetes. Selleks kasutati 600 erineval andmekogumil kanoonilist korrelatsioonianalüüsi ja klasterdamist, et leida mustreid elektroodide aktiivsusmõõdetes ja käitumuslike reaktsioonide keerukaid kombinatsioone kodeerituna inimaju signaalides. Lisaks kasutati masinõppemeetodeid, et analüüsida patsientide käitumissuundumusi riskivõtmise suhtes hasartmänguülesandes ja ennustada närviandmetest ülesandega seotud sündmusi nagu võitmine, kaotamine ja riskivõtmine. Enamiku meetoditega saavutati mõõdukad kuni head tulemused, kuid põhjalikum analüüs on siiski vajalik, et saada täielik arusaam sellest, kuidas orbitofrontaalse ajukoore aktiivsus põhjustab inimkäitumist otsuste tegemisel.
The orbitofrontal cortex (OFC) is a region sitting at the front of the brain which function is not fully understood. However, it has been clearly implicated in decision making as shown by many neuroimaging studies over the last decades. Recent work by Saez et al. [1] has found evidence that OFC activity of high frequency (HFA) between 70-200 Hz is directly related to behavioral responses during decision making tasks. In particular, Saez et al. showed that some modulations of HFA correlated with the human choice and outcome in a simple betting game. Saez et al. conducted their analysis with univariate linear regression, predicting HFA values from one task-related parameter at a time to find electrodes which encode decision making information. This Thesis focused on extending these results and analyses of Saez et al. by applying multivariate methods to discover complex signals and important patterns in the neural data. For this, canonical correlation analysis and biclustering were used on 600 different datasets to find evidence of patterns in electrode responses and complicated combinations of behavioral responses encoded in the human brain signals. In addition, machine learning methods were used to analyze the patients' behavioral tendencies towards risk-taking in a gambling task and to predict task-related events such as winning, losing and gambling from the neural data. Moderate to good performance was achieved with most methods, but in-depth analysis is still necessary to gain a full understanding of how activity in orbitofrontal cortex gives rise to human behavior in decision making tasks.
The orbitofrontal cortex (OFC) is a region sitting at the front of the brain which function is not fully understood. However, it has been clearly implicated in decision making as shown by many neuroimaging studies over the last decades. Recent work by Saez et al. [1] has found evidence that OFC activity of high frequency (HFA) between 70-200 Hz is directly related to behavioral responses during decision making tasks. In particular, Saez et al. showed that some modulations of HFA correlated with the human choice and outcome in a simple betting game. Saez et al. conducted their analysis with univariate linear regression, predicting HFA values from one task-related parameter at a time to find electrodes which encode decision making information. This Thesis focused on extending these results and analyses of Saez et al. by applying multivariate methods to discover complex signals and important patterns in the neural data. For this, canonical correlation analysis and biclustering were used on 600 different datasets to find evidence of patterns in electrode responses and complicated combinations of behavioral responses encoded in the human brain signals. In addition, machine learning methods were used to analyze the patients' behavioral tendencies towards risk-taking in a gambling task and to predict task-related events such as winning, losing and gambling from the neural data. Moderate to good performance was achieved with most methods, but in-depth analysis is still necessary to gain a full understanding of how activity in orbitofrontal cortex gives rise to human behavior in decision making tasks.