Klasifikasi Multikelas Kanker Payudara pada Citra Mamografi Menggunakan Multiview Convolutional Neural Network

Wiyoto, Dina Zhahrina Dwiputri (2026) Klasifikasi Multikelas Kanker Payudara pada Citra Mamografi Menggunakan Multiview Convolutional Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023221058-Undergraduate_Thesis (1).pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Kanker payudara merupakan salah satu penyebab utama kematian akibat kanker pada perempuan di Indonesia, sehingga deteksi dini melalui skrining mamografi menjadi komponen penting dalam penanganan klinis. Akan tetapi, interpretasi mamografi tetap merupakan tugas yang menantang: benign dan malignant dan normal dan benign dapat tampak serupa secara radiologis, kualitas citra bervariasi, dan penilaian terhadap suatu temuan kerap bersifat halus. Penelitian ini merancang dan mengevaluasi model klasifikasi multikelas untuk membedakan citra mamografi normal, benign, dan malignant menggunakan arsitektur Multiview Convolutional Neural Network (MVCNN) yang memanfaatkan dua tampilan utama mamografi, yaitu craniocaudal (CC) dan mediolateral oblique (MLO). Alur penelitian mencakup prapemrosesan citra (background removal, peningkatan citra berbasis CLAHE, standardisasi orientasi citra, dan pectoral muscle removal), ekstraksi fitur melalui dua cabang CNN paralel, penggabungan fitur multiview dengan strategi late fusion, penambahan modul attention (Squeeze-and-Excitation dan Spatial Attention), serta penerapan Grad-CAM untuk interpretasi visual terhadap keputusan model. Eksperimen dilakukan pada dataset Mini-DDSM dengan membandingkan arsitektur custom CNN yang dikembangkan secara bertahap terhadap pendekatan transfer learning menggunakan ResNet18. Model dengan performa terbaik, yaitu ResNet18 dengan konfigurasi full fine-tuning yang dikombinasikan dengan heavy data augmentation, mencapai akurasi 80%, presisi 80%, sensitivitas 81%, dan spesifisitas 90% pada tugas klasifikasi tiga kelas, dengan kesalahan klasifikasi paling banyak terjadi antara kelas normal dan benign akibat kemiripan visualnya yang tinggi. Hasil ini sejalan dengan studi tiga kelas yang sebanding dan menunjukkan bahwa pendekatan MVCNN dapat diterapkan pada klasifikasi mamografi multikelas, sebagai langkah awal menuju pengembangan sistem Computer-Aided Diagnosis (CAD) yang mendukung radiolog dalam deteksi dini kanker payudara.
======================================================================================================================================
Breast cancer is one of the leading causes of cancer-related mortality among women in Indonesia, making early detection through mammography screening an important component of clinical management. Mammogram interpretation, however, remains a demanding task benign and malignant lesions and benign and normal condition can appear radiologically similar, image quality varies, and the assessment of a finding is often subtle. This study designs and evaluates a multiclass classification model to distinguish normal, benign, and malignant mammographic images using a Multiview Convolutional Neural Network (MVCNN) architecture that leverages the two primary mammographic views, craniocaudal (CC) and mediolateral oblique (MLO). The workflow includes image preprocessing (background removal, CLAHE-based image enhancement, image orientation standardization, and pectoral-muscle removal), feature extraction through two parallel CNN branches, multiview feature fusion using a late-fusion strategy, the addition of attention modules (Squeeze-and-Excitation and Spatial Attention), and the application of Grad-CAM for visual interpretation of the model decisions. Experiments were conducted on the Mini-DDSM dataset, comparing a custom CNN architecture developed incrementally against a transfer-learning approach using ResNet18. The best-performing model, ResNet18 under a full fine-tuning configuration (all layers retrained) combined with heavy data augmentation, achieved 80% accuracy, 80% precision, 81% sensitivity, and 90% specificity on the three-class classification task, with most misclassifications occurring between the benign and normal classes due to their high visual similarity. These results are in line with a comparable three-class study and indicate that the MVCNN approach can be applied to multiclass mammogram classification, serving as an initial step toward a Computer-Aided Diagnosis (CAD) system that supports radiologists in the early detection of breast cancer.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	mamografi, klasifikasi multikelas, MVCNN, deep learning, Grad-CAM, CC, MLO; mammography, multiclass classification, MVCNN, deep learning, Grad-CAM, CC, MLO
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T59.7 Human-machine systems. 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:	Dina Zhahrina Dwiputri Wiyoto
Date Deposited:	04 Aug 2026 04:18
Last Modified:	04 Aug 2026 04:18
URI:	http://repository.its.ac.id/id/eprint/142435

Actions (login required)

View Item