Deteksi Kanker Mulut Berbasis Citra Menggunakan Algoritma Convolutional Neural Network

Arieffaza, Muhammad Naufal (2026) Deteksi Kanker Mulut Berbasis Citra Menggunakan Algoritma Convolutional Neural Network. Other thesis, Institute Teknologi Sepuluh Nopember.

Text 05111840000126-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (2MB)

Abstract

Deteksi dini kanker mulut sangat krusial untuk meningkatkan angka harapan hidup pasien, namun keterbatasan ahli patologi di daerah terpencil sering menjadi kendala utama, sehingga penelitian ini mengembangkan sistem klasifikasi citra histopatologi berbasis deep learning dengan membandingkan arsitektur Convolutional Neural Network (CNN) dan MobileNetV2 menggunakan dataset 1.528 citra yang terbagi seimbang ke dalam tiga kategori: Cancerous, Normal, dan Pre-cancerous. Empat model dibangun dalam studi ini, yaitu Custom CNN Baseline dan Custom CNN Improved yang dilengkapi BatchNormalization, Dropout 0,5, GlobalAveragePooling2D, serta Cosine Decay, serta MobileNetV2 Baseline dan MobileNetV2 Improved dengan fine-tuning dua tahap, yang kinerjanya diukur menggunakan akurasi dan F1-score. Dari hasil pengujian, Custom CNN Improved menunjukkan peningkatan akurasi yang cukup signifikan dari 64,71% menjadi 88,89% (naik sekitar 24,18%), sementara MobileNetV2 Improved berhasil mencatatkan akurasi tertinggi hingga 92,48% dengan F1-score macro 0,93, dengan kelas Pre-cancerous menjadi kategori paling sulit diidentifikasi diduga karena morfologi selnya memiliki kemiripan erat dengan sel normal. Sebagai luaran akhir, model MobileNetV2 Improved terbaik diintegrasikan ke dalam aplikasi berbasis web menggunakan Streamlit yang diharapkan menjadi alat bantu skrining awal praktis, terutama bagi fasilitas kesehatan dengan akses terbatas ke tenaga patologi ahli.
===========================================================================================================================
Early detection of oral cancer is crucial for improving patient survival rates, yet the scarcity of pathologists in remote areas often becomes a major obstacle, prompting this study to develop a deep learning-based histopathological image classification system by comparing Convolutional Neural Network (CNN) and MobileNetV2 architectures using a dataset of 1,528 balanced images divided into three categories: Cancerous, Normal, and Pre-cancerous. Four models were constructed in this study, namely Custom CNN Baseline and Custom CNN Improved equipped with BatchNormalization, Dropout 0.5, GlobalAveragePooling2D, and Cosine Decay, as well as MobileNetV2 Baseline and MobileNetV2 Improved with two-stage fine-tuning, whose performances were measured using accuracy and F1-score. From the experimental results, Custom CNN Improved demonstrated a significant accuracy improvement from 64.71% to 88.89% (an increase of approximately 24.18%), while MobileNetV2 Improved achieved the highest accuracy of 92.48% with a macro F1-score of 0.93, with the Pre-cancerous class being the most difficult category to identify, presumably due to its cellular morphology closely resembling normal cells. As the final output, the best MobileNetV2 Improved model was integrated into a web-based application using Streamlit, which is expected to serve as a practical early screening tool, particularly for healthcare facilities with limited access to expert pathologists.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	kanker mulut, deep learning, CNN, MobileNetV2, klasifikasi citra histopatologi. ============================================================ oral cancer, deep learning, CNN, MobileNetV2, histopathological image classification.
Subjects:	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) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Naufal Arieffaza
Date Deposited:	04 Aug 2026 01:21
Last Modified:	04 Aug 2026 01:21
URI:	http://repository.its.ac.id/id/eprint/142573

Actions (login required)

View Item