This doctoral research aims to explore the intersections and synergies between machine learning and neurology. The study's core is comprised of three distinct but interconnected investigations that highlight the potential of artificial intelligence (AI) in enhancing our understanding of neurodegenerative diseases and animal communication, as well as improving diagnostic methods. The first part of this research investigates the application of machine learning techniques in the diagnosis of Fronto-Temporal Dementia using Magnetic Resonance Imaging data. The study primarily uses a Multi-Voxel Pattern Analysis approach with Support Vector Machine and Random Forest algorithms for analysis. This section aims to address the challenges in the early detection of neurodegenerative diseases, providing medical professionals with a supportive diagnostic tool that could potentially improve treatment outcomes. The second part of the research delves into the analysis of ultrasonic vocalizations in mice, specifically focusing on the changes in ultrasonic communication patterns in mice treated with Cannabis sativa oil as compared to control mice. The investigation utilizes specialized recording equipment and dedicated software to analyze USVs, shedding light on the nuances of animal communication. This segment discusses the disparities in ultrasonic communication patterns between the two groups, correlating them with specific behaviors, presenting a comprehensive statistical exploration. The third part of the study presents a data processing pipeline to analyze mouse audio data, employing advanced signal processing techniques and machine learning. This segment introduces a meticulous data labeling system that assigns each audio segment to one of eight behavioral categories. Spectrogram computations are used to visualize distinct acoustic characteristics of high-frequency mouse vocalizations. Deep learning experiments undertaken in this part aim to uncover insights into the neural mechanisms underlying mouse ultrasonic vocalizations, thereby enriching our understanding of the intricate relationship between vocalization and behavior.
Questa ricerca di dottorato mira a esplorare le intersezioni e le sinergie tra l'apprendimento automatico e la neurologia. Il nucleo dello studio è costituito da tre indagini distinte ma interconnesse che mettono in evidenza il potenziale dell'intelligenza artificiale (IA) nel migliorare la nostra comprensione delle malattie neurodegenerative e della comunicazione animale, nonché nel migliorare i metodi di diagnosi. La prima parte di questa ricerca studia l'applicazione di tecniche di apprendimento automatico nella diagnosi della demenza frontotemporale utilizzando dati di risonanza magnetica. Lo studio utilizza principalmente un approccio di analisi a pattern multivoxel con algoritmi di Support Vector Machine e Random Forest per l'analisi. Questa sezione mira a affrontare le sfide nella rilevazione precoce delle malattie neurodegenerative, fornendo ai professionisti medici uno strumento diagnostico di supporto che potrebbe migliorare potenzialmente gli esiti del trattamento. La seconda parte della ricerca approfondisce l'analisi delle vocalizzazioni ultrasoniche nei topi, concentrandosi in particolare sulle modifiche nei modelli di comunicazione ultrasonica nei topi trattati con olio di Cannabis sativa rispetto ai topi di controllo. L'indagine utilizza attrezzature di registrazione specializzate e software dedicati per analizzare le vocalizzazioni ultrasoniche, facendo chiarezza sulle sfumature della comunicazione animale. Questo segmento discute delle disparità nei modelli di comunicazione ultrasonica tra i due gruppi, correlandoli a comportamenti specifici, presentando un'esaustiva esplorazione statistica. La terza parte dello studio presenta un flusso di elaborazione dei dati per analizzare le registrazioni audio dei topi, impiegando avanzate tecniche di elaborazione del segnale e di apprendimento automatico. Questo segmento introduce un meticoloso sistema di etichettatura dei dati che assegna ciascun segmento audio a una delle otto categorie comportamentali presenti nel dataset. I calcoli degli spettrogrammi vengono utilizzati per visualizzare le caratteristiche acustiche delle vocalizzazioni ad alta frequenza dei topi. Gli esperimenti di deep learning intrapresi in questa parte mirano a scoprire informazioni sui meccanismi neurali alla base delle vocalizzazioni ultrasoniche dei topi, arricchendo così la nostra comprensione della complessa relazione tra vocalizzazione e comportamento.
Researching the Potential of Artificial Intelligence to Support the Understanding of Neurological Diseases: The Cases for Frontotemporal Lobar Degeneration Detection and Mice Ultrasonic Communication Analysis / Pilipenko, Tatiana. - (2024 Apr 11).
Researching the Potential of Artificial Intelligence to Support the Understanding of Neurological Diseases: The Cases for Frontotemporal Lobar Degeneration Detection and Mice Ultrasonic Communication Analysis
PILIPENKO, TATIANA
2024-04-11
Abstract
This doctoral research aims to explore the intersections and synergies between machine learning and neurology. The study's core is comprised of three distinct but interconnected investigations that highlight the potential of artificial intelligence (AI) in enhancing our understanding of neurodegenerative diseases and animal communication, as well as improving diagnostic methods. The first part of this research investigates the application of machine learning techniques in the diagnosis of Fronto-Temporal Dementia using Magnetic Resonance Imaging data. The study primarily uses a Multi-Voxel Pattern Analysis approach with Support Vector Machine and Random Forest algorithms for analysis. This section aims to address the challenges in the early detection of neurodegenerative diseases, providing medical professionals with a supportive diagnostic tool that could potentially improve treatment outcomes. The second part of the research delves into the analysis of ultrasonic vocalizations in mice, specifically focusing on the changes in ultrasonic communication patterns in mice treated with Cannabis sativa oil as compared to control mice. The investigation utilizes specialized recording equipment and dedicated software to analyze USVs, shedding light on the nuances of animal communication. This segment discusses the disparities in ultrasonic communication patterns between the two groups, correlating them with specific behaviors, presenting a comprehensive statistical exploration. The third part of the study presents a data processing pipeline to analyze mouse audio data, employing advanced signal processing techniques and machine learning. This segment introduces a meticulous data labeling system that assigns each audio segment to one of eight behavioral categories. Spectrogram computations are used to visualize distinct acoustic characteristics of high-frequency mouse vocalizations. Deep learning experiments undertaken in this part aim to uncover insights into the neural mechanisms underlying mouse ultrasonic vocalizations, thereby enriching our understanding of the intricate relationship between vocalization and behavior.File | Dimensione | Formato | |
---|---|---|---|
Pilipenko_Thesis_30_12.pdf
accesso aperto
Descrizione: Tesi_Pilipenko_Tatiana
Tipologia:
Tesi di dottorato
Dimensione
5.01 MB
Formato
Adobe PDF
|
5.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.