Update Artikelilmiah: 51225

FILLIPUS ADITYA NUGROHO

NIM

Judul Artikel

Abstrak (Bhs. Indonesia)

Kanker serviks merupakan salah satu penyebab utama kematian pada perempuan, terutama di negara berkembang seperti Indonesia. Metode Inspeksi Visual Asam Asetat (IVA) banyak digunakan karena murah dan sederhana, namun hasilnya bergantung pada subjektivitas tenaga medis sehingga berisiko menimbulkan kesalahan diagnosis. Untuk meningkatkan objektivitas dan akurasi deteksi dini, penelitian ini mengusulkan pendekatan pengolahan citra digital melalui analisis fitur momen warna pada ruang warna RGB, YUV, dan LAB, serta kombinasi dengan fitur tekstur GLRLM, Tamura, dan LBP.
Penelitian dilakukan secara eksperimental menggunakan 162 citra kolposkopi pasca-IVA dari IARC Colposcopy Image Bank (75 abnormal, 87 normal). Tahapan meliputi prapemrosesan, ekstraksi fitur warna (mean, standar deviasi, skewness), dan tekstur (LBP, GLRLM, Tamura). Model diuji menggunakan algoritma XGBoost dan AdaBoost dengan evaluasi akurasi, presisi, spesifisitas, recall, dan F1-score serta analisis feature ranking.
Hasil menunjukkan bahwa model XGBoost dengan kombinasi fitur warna YUV dan tekstur memberikan performa terbaik dengan akurasi 90,91%, presisi 93,75%, spesifisitas 93,75%, recall 88,24%, dan F1-score 90,91%. Ruang warna YUV terbukti paling efektif karena mampu memisahkan luminansi dan krominansi secara optimal, sementara fitur tekstur berperan dominan dalam mendeteksi pola lesi. Kombinasi fitur warna dan tekstur secara keseluruhan meningkatkan akurasi dan stabilitas model klasifikasi.

Abtrak (Bhs. Inggris)

Cervical cancer is one of the leading causes of death among women, particularly in developing countries such as Indonesia. The Visual Inspection with Acetic Acid (VIA) method is widely used due to its affordability and simplicity; however, its results heavily depend on the subjectivity of medical personnel, which may lead to diagnostic errors. To enhance objectivity and accuracy in early detection, this study proposes a digital image processing approach through the analysis of color moment features in RGB, YUV, and LAB color spaces, combined with texture features including GLRLM, Tamura, and LBP. 
The study was conducted experimentally using 162 post-VIA colposcopy images from the IARC Colposcopy Image Bank, consisting of 75 abnormal and 87 normal images. The stages included preprocessing, extraction of color features (mean, standard deviation, skewness), and texture features (LBP, GLRLM, Tamura). The extracted features were evaluated using XGBoost and AdaBoost algorithms, with performance assessed through accuracy, precision, specificity, recall, and F1-score, as well as feature ranking analysis. 
The results show that the XGBoost model with a combination of YUV color features and texture features achieved the best performance, with 90.91% accuracy, 93.75% precision, 93.75% specificity, 88.24% recall, and 90.91% F1 score. The YUV color space proved to be the most effective, as it separates luminance and chrominance components efficiently, while texture features played a dominant role in detecting cervical lesion patterns. Overall, the combination of color and texture features enhanced the accuracy and stability of the classification model.

Kata kunci

Pembimbing 1

Pembimbing 2

Pembimbing 3

Tahun

Jumlah Halaman