Deteksi dan Visualisasi Berbasis Computer Vision Untuk Analisis Gambar Dermatologis Dalam Penilaian Keparahan Jerawat

Watef, Lulu'ul (2024) Deteksi dan Visualisasi Berbasis Computer Vision Untuk Analisis Gambar Dermatologis Dalam Penilaian Keparahan Jerawat. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 6026202007-Master_Thesis.pdf] Text
6026202007-Master_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 July 2026.

Download (1MB) | Request a copy

Abstract

Jerawat (acne vulgaris) merupakan salah satu penyakit paling umum dalam kehidupan sehari-hari utamanya pada masa remaja. Lebih dari 80% orang mengalami permasalahan jerawat pada masa remajanya. Dalam menilai tingkat keparahan jerawat perlu dilakukan penilaian secara obyektif dengan prosedur perawatan medis. Penilaian jerawat secara tradisional umumnya dilakukan oleh dokter kulit dengan menentukan jenis dan jumlah lesi dimana sering kali penilaian ini memakan waktu dan mengganggu efisiensi dalam penilaian jerawat. Kerangka kerja ini menguji praproses gambar dengan menggunakan teknik threshold, enhanced image, deteksi area merah dan penggabungan enhanced image dan deteksi area merah ke dalam klasifikasi jerawat dimana klasifikasi jerawat dibagai menjadi empat kelas mulai dari ringan, sedang, parah dan sangat parah. Dengan menggunakan beberapa metode klasifikasi seperti NN, SVM, KNN dan VGG16 serta diterapkannya fitur ekstraksi dihasilkan bahwasanya data masukan enhanced image dengan model klasifikasi menggunakan VGG16 memberikan hasil akurasi yang paling baik dengan tingkat akurasi mencapai 69%. Secara berturut-turut dengan menggunakan model klasifikasi VGG16 dengan data masukan threshold, deteksi area merah dan penggabungan enhanced dan area merah menghasilkan tingkat akurasi 54.42%, 52.99% dan 54.86%. Kedepannya metode praproses enhanced image dapat diterapkan pada kumpulan data yang berbeda dikarenakan tidak dipungkiri bahwasanya kumpulan data ACNE04 memiliki banyak noise pada citra.
====================================================================================================================================
Acne (acne vulgaris) is one of the most common diseases in everyday life, especially during adolescence. More than 80% of people experience acne problems during their teenage years. In assessing the severity of acne, it is necessary to assess it objectively with medical treatment procedures. Traditional acne assessments are generally carried out by dermatologists by determining the type and number of lesions, which often takes time and interferes with efficiency in acne assessment. This framework tests image preprocessing using threshold techniques, enhanced image, red area detection and combining enhanced image and red area detection into acne classification. Acne classification is divided into four classes ranging from mild, moderate, severe, and very severe. By using several classification methods such as NN, SVM, KNN and VGG16 and applying feature extraction, it resulted that the enhanced image input data with the classification model using VGG16 gave the best accuracy results with an accuracy level reaching 69%. Consecutively, using the VGG16 classification model using threshold input data, red area detection and combining enhanced and red areas produced accuracy levels of 54.42%, 52.99% and 54.86%. In the future, the enhanced image preprocessing method can be applied to different data sets because it cannot be denied that the ACNE04 data set has a lot of noise.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Acne Vulgaris, Classification, Image Preprocessing, Medical Image Neural Network, Threshold, Enhanced Image, Red Color Detection
Subjects: T Technology > T Technology (General) > T58.6 Management information systems
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User: Lulu'ul Watef
Date Deposited: 19 Feb 2024 06:49
Last Modified: 19 Feb 2024 06:49
URI: http://repository.its.ac.id/id/eprint/107545

Actions (login required)

View Item View Item