Text

Penentuan Hyper-Parameter Terbaik Model MobileNet-UNet untuk Segmentasi Kanker Kulit

---

ABSTRAK
Pembuatan fantom CTDI kepala dengan material alternatif telah dilakukan.
Fantom dibuat dengan metode sederhana menggunakan material resin polyester
dan methyl ethyl ketone peroxide (MEKP) telah dievaluasi pada penelitian ini.
Fantom CTDI kepala yang dikembangkan dibuat dengan berbagai variasi
komposisi volume material. Variasi yang digunakan merupakan rasio volume
resin dan MEKP. Pada penelitian ini variasi komposisi yang digunakan
diantaranya adalah 150:1, 200:1, 250:1 dan 300:1. Prosedur pada penelitian ini
dimulai dengan pembuatan fantom yang meliputi pencetakan dan pengeboran,
selanjutnya dilakukan pengukuran massa jenis dan pemeriksaan CT scan.
Fantom dibuat dengan dimensi silinder berdiameter 160 mm dan tinggi 150 mm.
Pada pengukuran massa jenis dilakukan penimbangan massa fantom dan
mengukur volume dimensi fantom. Untuk pengujian CT number fantom
digunakan CT scan dengan metode helical scanning. Nilai CT number yang
didapatkan dari masing-masing variasi fantom berbahan resin kemudian
dibandingkan dengan nilai CT number fantom CTDI standard. Hasil
perbandingan nilai CT number pada seluruh fantom menunjukkan nilai yang
relative sama. Nilai CT number dan standard deviasi dari fantom berbahan resin
bernilai 3-6% lebih tinggi dibandingkan dengan fantom standard berbahan
PMMA. Selain itu, pengujian statistik dengan uji p-value antara fantom
standard PMMA dan fantom resin telah dilakukan. Nilai probabilitas (p-value)
dibuat dengan batasan α=0.05, hasil uji yang diterima merupakan fantom dengan
komposisi 300:1. Hasil ini menunjukkan bahwa fantom tersebut paling mendekati
sama dengan fantom standard dan material resin polyester dengan MEKP sebagai
katalisator dapat dijadikan sebagai material alternatif fantom. Selain itu, salah satu
kelebihan resin polyester sebagai fantom adalah biaya produksi yang relatif
rendah.
Kata kunci: fantom CTDI kepala, resin polyester, CT number, CT scanviii
ABSTRACT
Synthesized of head CTDI phantom with alternative material have been
carried out. In this study, developed phantom made with a simple method using
polyester resin and methyl ethyl ketone peroxide (MEKP) material has been
evaluate. The CTDI phantom head developed was made with a variety of material
volume compositions. The variations used are volume ratio between resin and
MEKP. In this study, variations in the composition used 150: 1, 200: 1, 250: 1 and
300: 1. The procedure in this study began with making phantoms which include
printing and drilling, then the density measurements and CT scan measurements
were performed. Fantom is made with cylindrical dimensions of 160 mm in
diameter and 150 mm high. In measuring the density of mass carried out weighing
phantom mass and measuring the volume of phantom dimensions. For the
phantom CT number examination, CT scan is used by helical scanning method.
The comparisons of CT numbers measured in the polyester resin CTDI phantom
and the standard PMMA phantom in each composition variation showed that CT
numbers were relatively same. The CT number and standard deviation of the
developed phantoms were about 3-6% higher than the standard PMMA. Moreover,
statistical tests with p-value tests have been carried out. The probability value
(p-value) is made with α = 0.05, the test results achieved are phantoms with a
composition of 300: 1. This finding revealed that the phantom is almost same as
the standard phantom and polyester resin material with MEKP as a catalyst can be
used as an alternative material for phantom. It might be considered as an
alternative CTDI phantom for the hospitals do not own the standard PMMA
phantom. One advantage of the polyester resin CTDI phantom is its effective cost.
Kata kunci: head CTDI phantom, polyester resin, CT number, CT scan

ABSTRACT
Skin cancer, especially the type of melanoma is one type of deadly disease. The estimated 5-
year survival rate for melanoma is more than 99% if detected at the earliest stage, and to
around 14% if detected at the very last stage. So early detection is very important. Along
with the times, early detection of skin cancer can be aided with technology. Semantic
segmentation of skin lesions is one solution to help dermatologists improve the diagnosis of
skin cancer through dermoscopy images efficiently and accurately and can improve the
accuracy of classification of skin diseases. UNet is a semantic segmentation model that is
widely used in the field of medical image analysis. MobileNet is known for its ability to
overcome computational efficiency problems. This study uses the MobileNet-UNet model to
conduct semantic segmentation of skin cancer images by selecting the hyper-parameter
learning rate, epochs, image size, and the best pre-trained model. The dataset used is ISIC
2017 with the division of data into 2000 training data, 150 validation data, and 600 test
data. MobileNet-UNet is trained and tested with three scenarios. Of the three scenarios, the
highest Intersection over Union (IoU) score was 72.44% and the highest F1-Score was
83.33%. The results were obtained using a hyper-parameter learning rate of 0.001, epochs
at the best validation, image size 256 × 256, and using the ImageNet pre-trained model.
Keyword : Skin cancer, Semantic segmentation, MobileNet-UNet, Hyper-parameter

---

Ketersediaan

871F20III

871 F 20-ii

Perpustakaan FSM Undip (Referensi)

Tersedia

Informasi Detail

Judul Seri

ILMU KOMPUTER/INFORMATIKA

No. Panggil

871 F 20-ii

Penerbit

: ., 2020

Deskripsi Fisik

-

Bahasa

Indonesia

ISBN/ISSN

-

Klasifikasi

NONE

Tipe Isi

-

Tipe Media

-

Tipe Pembawa

-

Edisi

-

Subjek

-

Info Detail Spesifik

-

Pernyataan Tanggungjawab

Satriawan Rasyid Purnama

Versi lain/terkait

Lampiran Berkas

Komentar

---

Anda harus masuk sebelum memberikan komentar