Klasifikasi Citra Histopatologis Rongga Mulut Untuk Deteksi Oral Squamous Cell Carcinoma (OSCC) Berbasis Deep Learning

Irsyad, Dielian Maulana (2026) Klasifikasi Citra Histopatologis Rongga Mulut Untuk Deteksi Oral Squamous Cell Carcinoma (OSCC) Berbasis Deep Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221013-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Oral Squamous Cell Carcinoma (OSCC) merupakan salah satu jenis kanker rongga mulut yang memiliki angka mortalitas tinggi akibat keterlambatan diagnosis. Pemeriksaan histopatologi menggunakan pewarnaan Hematoxylin dan Eosin (H&E) masih menjadi gold standard dalam diagnosis OSCC, namun variasi warna yang muncul akibat perbedaan proses pewarnaan dapat memengaruhi konsistensi interpretasi citra serta performa sistem klasifikasi berbasis deep learning. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi citra histopatologis rongga mulut untuk deteksi OSCC dengan menerapkan Vahadane Stain Normalization dan membandingkan performa beberapa arsitektur Convolutional Neural Network (CNN).Dataset yang digunakan adalah Histopathological Imaging Database for Oral Cancer Analysis yang terdiri atas 1.224 citra histopatologis rongga mulut dengan perbesaran 100× dan 400×. Proses penelitian meliputi normalisasi warna menggunakan metode Vahadane dengan dua pendekatan pemilihan target reference, yaitu Manual Reference dan Mean RGB Dataset Reference, dilanjutkan dengan image resizing, augmentasi data, serta pelatihan model EfficientNet-B0, ResNet50, dan ResNet101 menggunakan skema Stratified 5-Fold Cross Validation. Evaluasi dilakukan menggunakan metrik accuracy, precision, sensitivity, specificity, Mean Sensitivity-Specificity (MSS), dan F1-score. Hasil penelitian menunjukkan bahwa perbesaran citra, metode pemilihan target reference, dan arsitektur CNN mempengaruhi performa klasifikasi. Konfigurasi terbaik diperoleh pada penggunaan Vahadane Stain Normalization dengan Manual Reference dan model EfficientNet-B0 pada dataset 100× yang menghasilkan accuracy 96,26 ± 1,85%, precision 98,15 ± 1,69%, sensitivity 95,21 ± 3,18%, specificity 97,62 ± 2,26%, MSS 96,41 ± 1,74%, dan F1-score 96,66 ± 2,21%. Hasil tersebut menunjukkan bahwa kombinasi normalisasi warna dan arsitektur CNN yang efisien mampu meningkatkan performa klasifikasi citra histopatologis OSCC serta berpotensi mendukung pengembangan sistem diagnosis berbantuan komputer untuk deteksi dini kanker rongga mulut.
========================================================================================================================
=============
Oral Squamous Cell Carcinoma (OSCC) is one of the most common oral malignancies and remains a major health concern due to its high mortality rate when diagnosed at advanced stages. Histopathological examination using Hematoxylin and Eosin (H&E) Staining is considered the gold standard for OSCC diagnosis. However, variations in Staining procedures can introduce color inconsistencies that affect both visual interpretation and the performance of deep learning models. This study aims to develop a deep learning-based classification system for oral histopathological images by applying Vahadane Stain Normalization and evaluating different Convolutional Neural Network (CNN) architectures for OSCC detection. The study utilized the Histopathological Imaging Database for Oral Cancer Analysis, consisting of 1,224 histopathological images with 100× and 400× magnifications. Vahadane Stain Normalization was performed using two target reference selection approaches, namely Manual Reference and Mean RGB Dataset Reference. Three CNN architectures, EfficientNet-B0, ResNet50, and ResNet101, were trained and evaluated using Stratified 5-Fold Cross Validation. Performance was assessed using accuracy, precision, sensitivity, specificity, Mean Sensitivity-Specificity (MSS), and F1-score. The results showed that image magnification, target reference selection, and CNN architecture significantly influenced classification performance. The best result was achieved using Vahadane Stain Normalization with a Manual Reference and EfficientNet-B0 on the 100× dataset, yielding an accuracy of 96.26 ± 1.85%, precision of 98.15 ± 1.69%, sensitivity of 95.21 ± 3.18%, specificity of 97.62 ± 2.26%, MSS of 96.41 ± 1.74%, and an F1-score of 96.66 ± 2.21%. These findings demonstrate that the proposed approach can effectively improve OSCC classification performance and has potential to support computer-aided diagnosis systems for early oral cancer detection.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	OSCC, histopatologi, Vahadane Stain Normalization, EfficientNet-B0, deep learning; OSCC, Histopathology, Vahadane Stain Normalization, EfficientNet-B0, Deep learning.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Dielian Maulana Irsyad
Date Deposited:	30 Jul 2026 00:47
Last Modified:	30 Jul 2026 00:47
URI:	http://repository.its.ac.id/id/eprint/140273

Actions (login required)

View Item