Klasifikasi Tumor Otak Pada Citra MRI Menggunakan en-CNN

Tjahyaningtijas, Hapsari Peni Agustin (2021) Klasifikasi Tumor Otak Pada Citra MRI Menggunakan en-CNN. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of DISERTASI_HapsariPeni_cetak.docx] Text
DISERTASI_HapsariPeni_cetak.docx - Accepted Version
Restricted to Repository staff only

Download (3MB) | Request a copy
[thumbnail of image49.png]
Preview
Image
image49.png

Download (95kB) | Preview
[thumbnail of image52.png]
Preview
Image
image52.png

Download (12kB) | Preview
[thumbnail of image27.png]
Preview
Image
image27.png

Download (10kB) | Preview
[thumbnail of image42.png]
Preview
Image
image42.png

Download (95kB) | Preview
[thumbnail of image7.png]
Preview
Image
image7.png

Download (64kB) | Preview
[thumbnail of image57.png]
Preview
Image
image57.png

Download (18kB) | Preview
[thumbnail of image30.png]
Preview
Image
image30.png

Download (24kB) | Preview
[thumbnail of image25.png]
Preview
Image
image25.png

Download (11kB) | Preview
[thumbnail of image51.png]
Preview
Image
image51.png

Download (188kB) | Preview
[thumbnail of image29.png]
Preview
Image
image29.png

Download (7kB) | Preview
[thumbnail of image23.png]
Preview
Image
image23.png

Download (6kB) | Preview
[thumbnail of image44.png]
Preview
Image
image44.png

Download (46kB) | Preview
[thumbnail of image48.png]
Preview
Image
image48.png

Download (35kB) | Preview
[thumbnail of image4.png]
Preview
Image
image4.png

Download (51kB) | Preview
[thumbnail of image60.png]
Preview
Image
image60.png

Download (16kB) | Preview
[thumbnail of image62.png]
Preview
Image
image62.png

Download (54kB) | Preview
[thumbnail of image33.png]
Preview
Image
image33.png

Download (11kB) | Preview
[thumbnail of equation] Other (equation)
image22.wmf - Other

Download (1kB)
[thumbnail of image10.png]
Preview
Image
image10.png

Download (62kB) | Preview
[thumbnail of image1.png]
Preview
Image
image1.png

Download (38kB) | Preview
[thumbnail of Equation] Text (Equation)
image11.wmf - Other

Download (1kB)
[thumbnail of image17.png]
Preview
Image
image17.png

Download (38kB) | Preview
[thumbnail of image28.png]
Preview
Image
image28.png

Download (10kB) | Preview
[thumbnail of image6.png]
Preview
Image
image6.png

Download (73kB) | Preview
[thumbnail of image63.png]
Preview
Image
image63.png

Download (115kB) | Preview
[thumbnail of image41.png]
Preview
Image
image41.png

Download (6kB) | Preview
[thumbnail of image31.png]
Preview
Image
image31.png

Download (28kB) | Preview
[thumbnail of image9.png]
Preview
Image
image9.png

Download (74kB) | Preview
[thumbnail of Equation] Text (Equation)
image12.wmf - Other

Download (954B)
[thumbnail of image3.png]
Preview
Image
image3.png

Download (142kB) | Preview
[thumbnail of image38.png]
Preview
Image
image38.png

Download (116kB) | Preview
[thumbnail of image71.png]
Preview
Image
image71.png

Download (78kB) | Preview
[thumbnail of Equation] Text (Equation)
image50.emf - Other

Download (26kB)
[thumbnail of image32.png]
Preview
Image
image32.png

Download (39kB) | Preview
[thumbnail of image61.png]
Preview
Image
image61.png

Download (31kB) | Preview
[thumbnail of image70.png]
Preview
Image
image70.png

Download (84kB) | Preview
[thumbnail of image35.png]
Preview
Image
image35.png

Download (43kB) | Preview
[thumbnail of image53.png]
Preview
Image
image53.png

Download (14kB) | Preview
[thumbnail of image15.png]
Preview
Image
image15.png

Download (2kB) | Preview
[thumbnail of Equation] Text (Equation)
image14.wmf - Other

Download (714B)
[thumbnail of image24.png]
Preview
Image
image24.png

Download (8kB) | Preview
[thumbnail of image20.png]
Preview
Image
image20.png

Download (12kB) | Preview
[thumbnail of image59.png]
Preview
Image
image59.png

Download (55kB) | Preview
[thumbnail of image40.png]
Preview
Image
image40.png

Download (35kB) | Preview
[thumbnail of image69.png]
Preview
Image
image69.png

Download (83kB) | Preview
[thumbnail of image45.png]
Preview
Image
image45.png

Download (40kB) | Preview
[thumbnail of image37.png]
Preview
Image
image37.png

Download (20kB) | Preview
[thumbnail of image54.png]
Preview
Image
image54.png

Download (111kB) | Preview
[thumbnail of image67.png]
Preview
Image
image67.png

Download (68kB) | Preview
[thumbnail of image56.png]
Preview
Image
image56.png

Download (21kB) | Preview
[thumbnail of Equation] Text (Equation)
image13.wmf - Other

Download (2kB)
[thumbnail of image39.png]
Preview
Image
image39.png

Download (71kB) | Preview
[thumbnail of image34.png]
Preview
Image
image34.png

Download (17kB) | Preview
[thumbnail of image65.png]
Preview
Image
image65.png

Download (21kB) | Preview
[thumbnail of image72.png]
Preview
Image
image72.png

Download (28kB) | Preview
[thumbnail of image47.png]
Preview
Image
image47.png

Download (72kB) | Preview
[thumbnail of image46.png]
Preview
Image
image46.png

Download (148kB) | Preview
[thumbnail of image58.png]
Preview
Image
image58.png

Download (62kB) | Preview
[thumbnail of image5.png]
Preview
Image
image5.png

Download (58kB) | Preview
[thumbnail of image66.png]
Preview
Image
image66.png

Download (66kB) | Preview
[thumbnail of image8.png]
Preview
Image
image8.png

Download (67kB) | Preview
[thumbnail of Equation] Text (Equation)
image73.wmf - Other

Download (67kB)
[thumbnail of image43.png]
Preview
Image
image43.png

Download (28kB) | Preview
[thumbnail of image21.png]
Preview
Image
image21.png

Download (11kB) | Preview
[thumbnail of image2.png]
Preview
Image
image2.png

Download (77kB) | Preview
[thumbnail of image64.png]
Preview
Image
image64.png

Download (47kB) | Preview
[thumbnail of image68.png]
Preview
Image
image68.png

Download (80kB) | Preview
[thumbnail of image19.png]
Preview
Image
image19.png

Download (13kB) | Preview
[thumbnail of image16.png]
Preview
Image
image16.png

Download (18kB) | Preview
[thumbnail of image36.png]
Preview
Image
image36.png

Download (40kB) | Preview
[thumbnail of image18.png]
Preview
Image
image18.png

Download (10kB) | Preview
[thumbnail of image26.png]
Preview
Image
image26.png

Download (10kB) | Preview
[thumbnail of image55.png]
Preview
Image
image55.png

Download (349kB) | Preview
[thumbnail of 07111760010006-Dissertation.docx] Text
07111760010006-Dissertation.docx
Restricted to Repository staff only

Download (4MB) | Request a copy
[thumbnail of 07111760010006-Dissertation.docx] Text
07111760010006-Dissertation.docx
Restricted to Repository staff only

Download (4MB) | Request a copy
[thumbnail of 07111760010006-Dissertation.docx] Text
07111760010006-Dissertation.docx
Restricted to Repository staff only

Download (4MB) | Request a copy
[thumbnail of 07111760010006-Dissertation.docx] Text
07111760010006-Dissertation.docx
Restricted to Repository staff only

Download (4MB) | Request a copy
[thumbnail of 07111760010006-Dissertation.pdf] Text
07111760010006-Dissertation.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2023.

Download (3MB) | Request a copy

Abstract

Tumor otak adalah salah satu penyakit yang paling umum terjadi pada sistem saraf pusat dan sifatnya berbahaya. Diagnosis dini sangat penting untuk perawatan pasien yang tepat. Klasifikasi biner tumor otak yang sering dicirikan dengan tumor otak ganas dan jinak yang melibatkan multi-sekuen MRI (T1, T2, T1CE, dan FLAIR), membuat pekerjaan ahli radiologi membosankan dan rawan terjadinya kesalahan. Pada penelitian ini, dikembangkan metode klasifikasi melalui tahap segmentasi dan metode klasifikasi langsung tanpa mealui tahap segmentasi untuk membantu proses klasifikasi tumor otak oleh ahli. Untuk metode klasifikasi melalui segmentasi, fokus penelitian terdapat pada pengembangan metode segmentasi otomatis untuk segmentasi tumor otak ganas yaitu Glioblastoma (GBM) dan tumor otak jinak yaitu Low Grade Glioma (LGG). Metode segmentasi dikembangkan menggunakan modifikasi U-Net. Arsitektur U-Net dievaluasi berdasarkan jumlah epoch dan nilai drop-out untuk mencapai arsitektur yang paling sesuai. Dari hasil eksperimen, model arsitektur yang paling sesuai untuk segmentasi tumor otak adalah arsitektur modifikasi U-Net atau mU-Net dengan jumlah epoch 90 dan nilai lapisan drop out 0,5. Hasil kinerja segmentasi ditunjukkan dengan nilai dice score sebesar 0,909 yang lebih besar dari penelitian sebelumnya. Metode segmentasi yang diusulkan mampu meningkatkan akurasi klasifikasi tumor otak sebesar 95,65% menggunakan DNN. Nilai akurasi tersebut 2,7% lebih tinggi dari pada jika menggunakan metode SVM yaitu sebesar 92,9%.
Dilain pihak, beberapa metode klasifikasi berdasarkan deep learning digunakan untuk mengklasifikasikan tumor otak. Performa masing-masing model sangat bergantung pada arsitektur CNN yang digunakan. Karena kompleksitas arsitektur CNN yang ada, penyetelan hyperparameter menjadi masalah dalam penerapannya. Pada penelitian ini diusulkan metode CNN yang disebut dengan en-CNN untuk mengatasi masalah ini. Metode ini didasarkan pada VGG-16 yang terdiri dari tujuh jaringan konvolusi, empat ReLU, dan empat max-pooling. Metode yang diusulkan digunakan untuk memfasilitasi penyetelan hyperparameter. Metode ini merupakan pendekatan dimana klasifikasi tumor otak dilakukan secara langsung tanpa terlebih dahulu melakukan proses segmentasi. Pendekatan baru terdiri dari tahapan berikut: preproses, augmentasi citra, dan penerapan metode en-CNN. Klasifikasi tumor otak dilakukan menggunakan empat sekuen MRI T1, T1CE, T2, dan FLAIR. Metode yang diusulkan memberikan akurasi pada dataset MRI multi-sekuen BraTS 2018 dengan akurasi 95,5% untuk T1, 95,5% untuk T1CE, 94% untuk T2, dan 97% untuk FLAIR dengan ukuran mini-batch 128 dan epoch 200 menggunakan fungsi optimasi ADAM. Akurasinya 4% lebih tinggi dari penelitian sebelumnya dalam dataset yang sama.
=====================================================================================================
Brain tumors are one of the most common diseases of the central nervous
system and are dangerous in nature. Early diagnosis is essential for proper patient
care. Radiologists need an automated system to identify brain tumor images. The
tumor identification process is a tedious and error-prone task. In addition, the binary
classification of brain tumors which are often characterized by malignant and
benign brain tumors involving multi-sequence MRI (T1, T2, T1CE, and FLAIR),
makes the work of radiologists quite challenging. In this study, a classification
method was developed through the segmentation stage. and the direct classification
method without going through the segmentation stage. For the classification method
through segmentation, the research focus is on the development of automatic
segmentation methods using U-Net modifications. The U-Net architecture was
evaluated based on the number of epochs and drop-out values to achieve the most
suitable architecture for automatic segmentation of glioblastoma brain tumors.
From the experimental results, the most suitable architectural model for brain tumor
segmentation is the mU-Net architecture with 90 epochs and a dropout layer value
of 0.5. The results of segmentation performance are indicated by a dice score of
0.909, which is greater than the previous study. Using DNN, the proposed
segmentation method can improve the accuracy of brain tumor classification by
95.65%. The accuracy value is 2.7 % higher than 92.9 % when using the SVM
method.
On the other hand, several classification methods based on deep learning are
used to classify brain tumors. The performance of each model is highly dependent
on the CNN architecture used. Due to the complexity of the existing CNN
architecture, hyperparameter tuning is a problem in its implementation. In this
study, a CNN method called en-CNN is proposed to overcome this problem. This
method is based on VGG-16 which consists of seven convolution networks, four
ReLUs, and four max-poolings. The proposed method is used to facilitate
hyperparameter tuning. This method is an approach where the classification of brain
tumors is done directly without first doing the segmentation process. The new
approach consists of the following stages: preprocessing, image augmentation, and
application of the en-CNN method. Brain tumor classification was performed using
four MRI sequences T1, T1CE, T2, and FLAIR. The proposed method provides an
accuracy of the 2018 BraTS multi-sequence MRI dataset with an accuracy of 95.5%
for T1, 95.5% for T1CE, 94% for T2, and 97% for FLAIR with mini-batch sizes of
128 and epoch 200 using the function ADAM optimization. The accuracy is 4%
higher than previous studies in the same dataset

Item Type: Thesis (Doctoral)
Additional Information: -
Uncontrolled Keywords: Segmentasi tumor otak, U-Net, drop-out, dice score, hyperparameter, Brain tumor segmentation, U-Net, drop-out, dice score, hyperparameter
Subjects: T Technology > T Technology (General) > T57.5 Data Processing
T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User: Hapsari Peni Agustin Tjahyaningtijas
Date Deposited: 17 Aug 2021 06:56
Last Modified: 17 Aug 2021 06:57
URI: http://repository.its.ac.id/id/eprint/87168

Actions (login required)

View Item View Item