Deep Learning untuk Diagnosis Gangguan Psikiatris dari Citra Magnetic Resonance Otak dengan Augmentasi Data Berbasis Generative Adversarial Network

Masengi, Gracia Angelina Jeniffer (2023) Deep Learning untuk Diagnosis Gangguan Psikiatris dari Citra Magnetic Resonance Otak dengan Augmentasi Data Berbasis Generative Adversarial Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 07311940000053-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (5MB) | Request a copy

Abstract

Belum adanya ukuran biomarker objektif dalam diagnosis merupakan salah satu tantangan yang dihadapi bidang psikiatri. Diagnosis umumnya masih dilakukan secara tradisional berdasarkan gejala perilaku. Kemajuan dalam bidang neuroimaging mendorong berbagai penelitian untuk menelusuri penanda biologis gangguan psikiatris dari citra otak. Di antaranya pemanfaatan deep learning untuk klasifikasi menunjukkan potensi yang besar. Dalam analisis citra medis, deep learning juga semakin diteliti pemanfaatannya untuk mengatasi masalah kurangnya sampel data dan ketidakseimbangan kelas, melalui model generatif seperti generative adversarial network (GAN). Pada penelitian ini, diimplementasikan metode diagnosis gangguan psikiatris dari citra magnetic resonance otak secara otomatis dengan deep learning. Diagnosis dilakukan melalui penerapan convolutional neural network (CNN) untuk klasifikasi citra otak ke dalam kelas: (1) Sehat, (2) Skizofrenia, dan (3) Gangguan Bipolar. Untuk menangani ketidakseimbangan kelas dan jumlah data yang sedikit, dimanfaatkan sintesis citra melalui jaringan generative adversarial network sebagai augmentasi data. Metode yang diajukan diharapkan dapat memajukkan state-of-the-art diagnosis bidang psikiatris dan mendorong penelitian lebih lanjut akan pemanfaatan pencitraan otak dalam skema diagnostik. Hasil implementasi menemukan bahwa jumlah data dan variabilitas subjek memiliki pengaruh yang besar dalam keberhasilan metode diagnosis gangguan psikiatris. Pada penelitian ini, augmentasi berbasis GAN belum berhasil untuk menutupi masalah kekurangan data dalam pembelajaran model CNN. Disimpulkan bahwa untuk membangun sistem diagnosis otomatis gangguan psikiatris yang efektif, penelitian perlu mengumpulkan jumlah data yang memadai dengan memperhatikan variabilitas pada subjek.
===============================================================================================================================
The lack of objective biomarker measures in diagnosis is one of the challenges faced in the field of psychiatry. Diagnosis is generally still based on traditional methods relying on behavioral symptoms. Advances in neuroimaging have led to various studies exploring biological markers of psychiatric disorders from brain images. Among these, the utilization of deep learning for classification shows great potential. In medical image analysis, deep learning is also being increasingly researched for addressing the issues of limited data samples and class imbalance, through generative models like generative adversarial network (GAN). This study implements a method for automatically diagnosing psychiatric disorders from magnetic resonance brain images using deep learning. The diagnosis is performed by utilizing convolutional neural network (CNN) to classify brain images into classes: (1) Healthy, (2) Schizophrenia, and (3) Bipolar Disorder. To address class imbalance and limited data, image synthesis is performed using a generative adversarial network as data augmentation. The proposed method aims to improve the state-of-the-art in psychiatric diagnosis and encourage further research on the utilization of brain imaging in diagnostic schemes. The implementation results reveal that data quantity and subject variability has a significant impact on the success of psychiatric disorder diagnosis methods. In this study, GAN-based augmentation did not effectively overcome the data scarcity issue in training the CNN model. It is concluded that to build an effective automatic psychiatric disorder diagnosis system, collection of adequate data samples needs to be ascertained while considering the effect of subject variability.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Automatic Diagnosis, Psychiatric Disorders, Diagnosis Otomatis, Gangguan Psikiatris, Deep Learning, Magnetic Resonance Imaging (MRI), Generative Adversarial Network (GAN)
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques T Technology > TA Engineering (General). Civil engineering (General) > TA174 Computer-aided design.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Gracia Angelina Jeniffer Masengi
Date Deposited:	24 Aug 2023 07:08
Last Modified:	24 Aug 2023 07:08
URI:	http://repository.its.ac.id/id/eprint/101823

Actions (login required)

View Item