Spatially Constrained Neo-Normal Mixture Model (SC-Nenomimo) Dengan Pendekatan Bayesian Pada Segmentasi Citra MRI Tumor Otak

Pravitasari, Anindya Apriliyanti (2020) Spatially Constrained Neo-Normal Mixture Model (SC-Nenomimo) Dengan Pendekatan Bayesian Pada Segmentasi Citra MRI Tumor Otak. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 06211560010003-Disertation.pdf] Text
06211560010003-Disertation.pdf - Accepted Version
Restricted to Repository staff only

Download (53MB) | Request a copy

Abstract

Segmentasi citra MRI dalam dunia medis bertujuan untuk memisahkan Region of Interest (ROI) atau segmen yang dianggap penting secara medis dengan segmen-segmen lainnya dalam citra. Hasil dari segmentasi citra MRI, dapat digunakan oleh dokter dan tenaga medis untuk mendiagnosa letak dan batas, tumor pada ROI guna mengambil keputusan medis..
Beberapa metode untuk segmentasi dikembangkan dalam analisis cluster, salah satunya adalah Model-based Clustering. Beberapa model yang telah dikembangkan masih menggunakan distribusi Normal sebagai distribusi pembangun model mixture-nya. Hal ini kurang cocok dengan citra MRI yang pola datanya tidak selalu simetris. Distribusi yang diusulkan dalam penelitian ini adalah Neo-Normal, yaitu distribusi adaptif yang dapat merelaksasi sifat distribusi Normal, namun dapat pula mengakomodasi pola asimetris, landai maupun runcing. Model yang terbentuk adalah Neo-Normal Mixture Model (Nenomimo).
Kelemahan pada model yang menggunakan distribusi Normal adalah bahwa antar pixel dalam citra masih dianggap independen, sehingga model akan lebih peka terhadap noise. Untuk mengatasi noise, model Nenomimo digabungkan dengan metode pengidentifikasi lokasi pixel, yaitu Markov Random Fields, sehingga nama modelnya menjadi Spatially constrained Neo-Normal Mixture Model (Sc-Nenomimo). Pendekatan Bayesian digunakan dalam estimasi parameter model, karena telah terdapat informasi awal distribusi prior. Implementasi dari model digunakan untuk segmentasi citra MRI tumor otak dan hasilnya direstrukturisasi kedalam bentuk cita 3D untuk memberikan visualisasi yang lebih baik.
Hasil segmentasi dengan Nenomimo dan Sc-Nenomimo memiliki tingkat akurasi yang baik, hal ini diberikan oleh nilai Misclassification Ratio yang kurang dari 2%. Pola citra MRI dapat didekati dengan baik oleh kedua model dengan nilai Fit Distribution Ratio sekitar 65%. Struktur 3D sangat membantu mengetahui letak dan batas tumor otak. Selain itu dengan struktur 3D dapat diperoleh estimasi volume tumor, yaitu sekitar 1192 s/d 1572 mm3.
=======================================================================================================
Medical image segmentation aims to separate the tumor area as the Region of Interest from the other segment in MRI brain tumor image. The segmentation results will provide some information such as the location and also give a clear boundary of the tumor. This will help the doctor to run the safe surgical treatment and minimize the damage potential of the healthy part of the brain. Image segmentation through clustering analysis has been widely developed under several algorithms. This study chooses the model-based clustering in the form of the finite mixture model since it could cluster the observation by considering the data pattern.
The previous state of the art uses the Gaussian distribution to construct the model. This model has disadvantage since the symmetrical property of Gaussian is not always fit the MRI data pattern. To solve this problem, this study uses Neo-Normal distribution to replace the Gaussian. The Neo-Normal is a relaxation of Normal distribution which can accommodate both symmetrical and asymmetrical patterns along with the ability to have long and fat tail properties. The proposed model is called Neo-Normal mixture model or Nenomimo.
The other disadvantage of Gaussian model is the assumption of pixel independencies. This assumption makes noise would be group in the same cluster as the tumor. To overcome this problem, this study tries to hybrid the Nenomimo with the Markov Random Field. The model then becomes a Spatially constraint Neo-Normal mixture model (Sc-Nenomimo). Bayesian coupled with Markov chain Monte Carlo uses in the optimization since there is prior information in every parameter distribution.
This study has succeeded to provide the Nenomimo and Sc-Nenomimo with the application of MRI based brain tumor segmentation. The segmentation results for both models have high accuracy since the Misclassification Ratio is less than 2%. Fit Distribution Ratio claim that both models could represent the MRI pattern of about 65%. In addition, the segmented images constructed in a 3D structure to give better visualization. With this structure, the location and boundary of the tumor area are more clear to see, and we can obtain the tumor volume estimation of about 1192 1572 mm3.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: image segmentation, clustering, Bayesian, Neo-Normal Mixture Model, Markov Random Fields
Subjects: H Social Sciences > HA Statistics > HA30.6 Spatial analysis
Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory.
T Technology > T Technology (General) > T57.5 Data Processing
T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
T Technology > TA Engineering (General). Civil engineering (General) > TA167.5 Neurotechnology. Neuroadaptive systems
Divisions: Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49001-(S3) PhD Thesis
Depositing User: Anindya Apriliyanti Pravitasari
Date Deposited: 18 Dec 2020 04:16
Last Modified: 18 Dec 2020 04:16
URI: http://repository.its.ac.id/id/eprint/82309

Actions (login required)

View Item View Item