FMRI kullanılarak otomatik DEHB teşhisi
Loading...

Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Open Access Color
OpenAIRE Downloads
OpenAIRE Views
Abstract
Dikkat eksikliği hiperaktivite bozukluğu (DEHB), birçok çocuk ve ergenin okul ve sosyal yaşamını olumsuz yönde etkilemektedir. DEHB aynı zamanda ailelere ve topluma da önemli yükler getirebilmektedir. Ayrıca, günümüzdeki dijitalleşme düzeyinin dikkat sürelerine olumsuz etki ettiği, belirlenmiş oyun alanlarının azalmasının ise çocuklarda biriken enerjinin dışarı atılmasında olumsuz etki yarattığı gözlenmektedir. Son yıllarda bu faktörlerin, en bilinen ruhsal sağlık bozukluklarından biri olan bu hastalığın farkındalığına katkıda bulunduğu düşünülmektedir. Görüntüleme ve makine/derin öğrenmedeki hızlı gelişmeler sayesinde, teşhisi acil ve önemli olan DEHB için geçmişte geleneksel yöntemlerle kullanılanlardan daha güvenilir teşhis yöntemleri artık mevcuttur. Bu tez çalışmasında oldukça geniş bir literatür taraması gerçekleştirilmiştir. Literatürde gözlenen bazı eksikler de gözetilerek, son zamanlarda DEHB teşhisinde sıklıkla kullanılan fonksiyonel manyetik rezonans görüntüleme (fMRG) verilerinden faydalanılarak bazı alternatif yöntem önerileri sunulmuştur. Bu alternatif yöntem önerilerinin etkinliğinin niceliksel olarak karşılaştırılabilmesi amacıyla DEHB teşhisinde genelde kullanılan yöntemleri içeren, sıklıkla atıf alan ve karşılaştırma için sıklıkla kullanılan 3 adet çalışma seçilmiştir. Bu çalışmalarda kullanılan yöntemler ortaklaştırılarak, bir referans modeli oluşturulmuştur. Bu referans modelin parametre optimizayonundan sonra alternatif yöntem önerileri tek tek ve bir arada referans modele entegre edilmiştir. Her bir entegrasyon sonucunda oluşan yapı optimize edilerek referans model ve literatür ile niceliksel olarak karşılaştırılmıştır. DEHB teşhisi için oluşturulan modellerin genelde sahip olduğu her bir aşama için ayrı bir alternatif yöntem önerisi yapılmıştır. Tüm beyin bölgelerinin kullanımı yerine varsayılan mod ağı (VMA) beyin bölgelerinin kullanılması, beyin bölgeleri arasındaki işlevsel bağlantıların hesaplanması için Pearson'ın ilişki katsayısı yöntemi yerine dalgacık dönüşüm uyumu (DDU) yönteminin kullanılması ve veri dengeleme ile özellik seçimi için sentetik azınlık aşırı örnekleme tekniği ve elastik ağ tabanlı özellik seçici yerine oto kodlayıcı'nın (OK) kullanılması önerilmiştir. Önerilen yöntemlerle elde edilen yapıların hepsi DEHB teşhisinde gelişim göstermiştir. Önerilen yöntemlerin referans modele entegre edilmesi ile oluşan yapılar arasında en yüksek sınıflandırma sonucuna ulaşan yapının VMA ve OK'nin birlikte kullanıldığı yapı olduğu gözlendiği için bu yapı geliştirilmiştir. Bunun için literatürde sıklıkla araştırılan teknik yöntemler yerine DEHB ile beyin bölgelerinin ilişkisi araştırılmıştır. Literatürde bu alandaki boşluk da gözetilerek DEHB'nin bireylerin hareket, dürtüsellik, hissi and karar alma işlevlerinde hasarlar oluşturmasından hareketle beynin 'basal ganglia', 'limbic system', and 'frontal cortex' alt sistemlerine odaklanılmıştır. Bu alt sistemleri oluşturan beyin bölgeleri seçilerek DEHB teşhisinde tüm beyin bölgelerinin kullanılmasından sonra en çok tercih edilen ağ olan VMA beyin bölgelerinin kullanılmasına bir alternatif oluşturulması istenmiştir. Bu alternatif oluşturulurken DEHB teşhisinde özelleşmiş bir odaklanma ve gelişimin elde edilmesi amaçlanmıştır. Bu uygulama esnasında bu sefer DEHB alt tipleri de sınıflandırmaya dahil edilmiş ve hemen tüm sınıflandırma olasılıkları için uygulama tekrarlanarak alternatif yöntem önerilerindeki gibi geniş, karşılaştırmalı ve niceliksel sonuçlar elde edilmiştir. Bu sonuçlardan gözlendiği üzere DEHB teşhisi için doğru beyin bölgeleri seçilmiştir. Ayrıca niceliksel sonuçlardan gözlendiği üzere bu tez çalışmasında DEHB-H teşhisine dair olası bir sorun üzerine bir hipotez ortaya atılmış ve niceliksel sonuçlarla incelenmiştir. Literatürde kullanılan veri tabanlarındaki DEHB-H teşhisi konan hasta sayısındaki dramatik azlık ve DEHB teşhisinde kullanılan öznel kriterlerdeki yoruma açıklık, bu çalışmada elde edilen geniş, karşılaştırmalı ve niceliksel sonuçlarla desteklenmiştir.
Attention deficit hyperactivity disorder (ADHD) has a negative impact on the school and social lives of many children and adolescents. ADHD can also impose considerable burdens on families and society. In addition, the contemporary level of digitalisation has been demonstrated to have a detrimental effect on attention spans, whilst the decrease in designated play areas has been shown to have a negative impact on the release of pent-up energy in children. In recent years, it is believed that these factors may have contributed to an extrally increased prevalence of this disorder, which is already one of the most well-known mental health disorders. Thanks to rapid developments in imaging and machine/deep learning, more reliable diagnostic methods for the urgent need of ADHD diagnosis are now available than those used in the past in a traditional way. A comprehensive review of the existing literature was conducted as a fundamental component of the thesis study. In light of certain deficiencies identified within the extant literature, a number of alternative methodology proposals have been advanced, drawing upon the utilisation of functional magnetic resonance imaging (fMRI) data, which has been frequently used in the diagnosis of ADHD recently. In order to quantitatively compare the effectiveness of these alternative method suggestions, three frequently cited and frequently used studies to compare the results in the literature were selected for comparison, which included methods commonly used in the diagnosis of ADHD. The methodologies employed in these studies were combined, thus yielding a reference model. Following parameter optimization of the reference model, the proposed alternative methods were integrated into the reference model both individually and together. The resulting structure from each integration was optimized and then quantitatively compared with the reference model and existing literature. A separate alternative method proposal has been made for each stage that the structures created for the diagnosis of ADHD generally have. It is proposed to integrate default mode network (DMN) brain regions instead of using whole brain regions, to use wavelet transform coherence (WTC) instead of Pearson's correlation coefficient for calculating functional connectivites between brain regions, and to use autoencoder (AE) instead of synthetic minority oversampling technique and elastic net-based feature selector for data balancing and feature selection. All of the structures obtained with the proposed methods demonstrated an improvement in the diagnosis of ADHD. The development of this model was driven by the observation that the structure that yielded the highest classification result among the structures formed by integrating the proposed methods into the reference model was the structure in which DMN and AE were integrated together. In order to address the gap in the literature, the present study investigated the relationship between ADHD and brain regions, as opposed to the technical methods that are frequently researched in the existing literature on the subject. Considering the gap in the literature in this area, the focus was on the basal ganglia, limbic system, and frontal cortex subsystems of the brain, considering that ADHD impairs individuals' motor, impulse, emotional, and decision-making functions. By selecting the brain regions that constitute these subsystems, it was desired to propose an alternative to the use of DMN brain regions, which is the most used network after the use of all brain regions in the diagnosis of ADHD. The objective of proposing this alternative was to achieve a specialised focus and improvement in the diagnosis of ADHD. During this application, ADHD subtypes were included in the classification process, and the application was repeated for nearly all classification possibilities. This resulted in the attainment of comprehensive, comparative, and quantitative results, as was the case in the proposed alternative methods. As observed from these results, the appropriate brain regions were identified for the diagnosis of ADHD. Furthermore, as evidenced by the quantitative results, a hypothesis concerning a potential issue in the diagnosis of ADHD-H was proposed and examined with the quantitative results in the thesis study. The dramatic paucity of participants diagnosed with ADHD-H in the databases utilised in the extant literature, coupled with the inherent ambiguity in the subjective criteria employed in traditional ADHD diagnosis, is substantiated by the comprehensive, comparative, and quantitative results presented in the study.
Attention deficit hyperactivity disorder (ADHD) has a negative impact on the school and social lives of many children and adolescents. ADHD can also impose considerable burdens on families and society. In addition, the contemporary level of digitalisation has been demonstrated to have a detrimental effect on attention spans, whilst the decrease in designated play areas has been shown to have a negative impact on the release of pent-up energy in children. In recent years, it is believed that these factors may have contributed to an extrally increased prevalence of this disorder, which is already one of the most well-known mental health disorders. Thanks to rapid developments in imaging and machine/deep learning, more reliable diagnostic methods for the urgent need of ADHD diagnosis are now available than those used in the past in a traditional way. A comprehensive review of the existing literature was conducted as a fundamental component of the thesis study. In light of certain deficiencies identified within the extant literature, a number of alternative methodology proposals have been advanced, drawing upon the utilisation of functional magnetic resonance imaging (fMRI) data, which has been frequently used in the diagnosis of ADHD recently. In order to quantitatively compare the effectiveness of these alternative method suggestions, three frequently cited and frequently used studies to compare the results in the literature were selected for comparison, which included methods commonly used in the diagnosis of ADHD. The methodologies employed in these studies were combined, thus yielding a reference model. Following parameter optimization of the reference model, the proposed alternative methods were integrated into the reference model both individually and together. The resulting structure from each integration was optimized and then quantitatively compared with the reference model and existing literature. A separate alternative method proposal has been made for each stage that the structures created for the diagnosis of ADHD generally have. It is proposed to integrate default mode network (DMN) brain regions instead of using whole brain regions, to use wavelet transform coherence (WTC) instead of Pearson's correlation coefficient for calculating functional connectivites between brain regions, and to use autoencoder (AE) instead of synthetic minority oversampling technique and elastic net-based feature selector for data balancing and feature selection. All of the structures obtained with the proposed methods demonstrated an improvement in the diagnosis of ADHD. The development of this model was driven by the observation that the structure that yielded the highest classification result among the structures formed by integrating the proposed methods into the reference model was the structure in which DMN and AE were integrated together. In order to address the gap in the literature, the present study investigated the relationship between ADHD and brain regions, as opposed to the technical methods that are frequently researched in the existing literature on the subject. Considering the gap in the literature in this area, the focus was on the basal ganglia, limbic system, and frontal cortex subsystems of the brain, considering that ADHD impairs individuals' motor, impulse, emotional, and decision-making functions. By selecting the brain regions that constitute these subsystems, it was desired to propose an alternative to the use of DMN brain regions, which is the most used network after the use of all brain regions in the diagnosis of ADHD. The objective of proposing this alternative was to achieve a specialised focus and improvement in the diagnosis of ADHD. During this application, ADHD subtypes were included in the classification process, and the application was repeated for nearly all classification possibilities. This resulted in the attainment of comprehensive, comparative, and quantitative results, as was the case in the proposed alternative methods. As observed from these results, the appropriate brain regions were identified for the diagnosis of ADHD. Furthermore, as evidenced by the quantitative results, a hypothesis concerning a potential issue in the diagnosis of ADHD-H was proposed and examined with the quantitative results in the thesis study. The dramatic paucity of participants diagnosed with ADHD-H in the databases utilised in the extant literature, coupled with the inherent ambiguity in the subjective criteria employed in traditional ADHD diagnosis, is substantiated by the comprehensive, comparative, and quantitative results presented in the study.
Description
Keywords
Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
Turkish CoHE Thesis Center URL
Fields of Science
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
196
