Klasifikasi Tumor Payudara dengan Infrared Thermography berbasis Convolutional Neural Network (CNN)

Rosita, Afifah Hasnia Nur (2026) Klasifikasi Tumor Payudara dengan Infrared Thermography berbasis Convolutional Neural Network (CNN). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211007-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Kanker payudara merupakan salah satu kanker dengan prevalensi tertinggi di dunia dan menyumbang sekitar 2,3 juta kasus baru dan 670.000 kematian per tahun. Meskipun mammmografi merupakan standar emas dari skrining awal kanker payudara, sensitivitasnya menurun pada payudara padat sehingga meningkatkan risiko false‑negatives. Di sisi lain, Pemeriksaan payudara sendiri (SADARI) memiliki sensivitas rendah (20-30%) dan bergantung pada kemampuan subjektif. Penelitian ini mengusulkan klasifikasi dua kelas (benign vs malignant) menggunakan infrared thermography (IRT) sebagai modalitas pendukung yang terjangkau, portabel, bebas radiasi, dan peka terhadap perubahan metabolik. Dataset Mendeley Breast Thermography (88 benign, 68 malignant) disegmentasi menjadi 340 ROI payudara tunggal, kemudian dikonversi ke JET colormap untuk meningkatkan kontras dan dinormalisasi orientasinya agar seluruh ROI menghadap kanan. Augmentasi pada level ROI menghasilkan 3.465 citra (1.735 benign, 1.730 malignant) melalui variasi rotasi kecil, zoom, kombinasi rotasi‑zoom, elastic deformation, bilateral smoothing, dan penambahan thermal noise. Model klasifikasi menggunakan VGG‑16 pre‑trained ImageNet dengan fine‑tuning, di mana blok konvolusional 1–4 dibekukan, blok 5 dan classification head tetap dilatih, dan GlobalAveragePooling2D menggantikan flattening untuk mereduksi parameter padat. Model mencapai akurasi 90,54%, precision 91,24%, sensitivity 89,45%, specificity 91,60%, dan AUC 97,25%. Temuan ini menunjukkan bahwa kombinasi IRT, transfer learning, dan augmentasi terarah berpotensi menjadi modalitas pendukung yang efektif untuk melengkapi mammografi dalam screening dini tumor payudara.
====================================================================================================================================
Breast cancer is one of the most prevalent cancers worldwide, accounting for approximately 2.3 million cases and 670,000 deaths globally. Mammography, the current gold standard for breast cancer screening, has reduced sensitivity in women with dense breast tissue, leading to higher false-negative rates. Additionally, clinical breast self-examination (SADARI) is limited by low sensitivity (20-30%) and operator-dependent variability. This study proposes infrared thermography (IRT) as a complementary approach for breast tumor classification, offering better accessibility, operational efficiency, and absence of ionizing radiation. The Mendeley Breast Thermography dataset (88 benign, 68 malignant cases) was segmented into 340 bilateral regions of interest (ROI) using ImageJ. Preprocessing included JET colormap conversion to enhance thermal contrast and horizontal flipping for spatial normalization. Multi-level ROI-based augmentation produced 3,465 training images (1,735 benign, 1,730 malignant) using six techniques: small angle rotation, zoom, combined rotation-zoom, elastic deformation, bilateral smoothing, and thermal noise injection. The classification model employed pre trained VGG-16 with selective fine-tuning of convolutional block 5 and the classification head, utilizing regularization strategies including dropout, L2 weight decay, BatchNormalization, and GlobalAveragePooling2D. The model achieved 90.54% accuracy, 91.24% precision, 89.45% sensitivity, 91.60% specificity, and 97.25% AUC. These results demonstrate that IRT combined with transfer learning and targeted augmentation provides accurate thermal abnormality detection, supporting its role as an effective adjunct to mammography for early breast tumor screening.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	deep learning, kanker payudara, infrared thermography, convolutional neural network (CNN), VGG-16 deep learning, breast cancer, infrared thermography, convolutional neural network (CNN), VGG-16
Subjects:	R Medicine > RB Pathology
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Afifah Hasnia Nur Rosita
Date Deposited:	02 Feb 2026 03:25
Last Modified:	02 Feb 2026 03:28
URI:	http://repository.its.ac.id/id/eprint/131448

Actions (login required)

View Item