Deteksi Sel Pada Citra Imunohistokimia Menggunakan Deep Learning

Suciningtyas, Laras (2026) Deteksi Sel Pada Citra Imunohistokimia Menggunakan Deep Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022241014-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (722kB) | Request a copy

Abstract

Kanker payudara merupakan jenis kanker dengan jumlah kasus tertinggi di Indonesia, dengan tingkat deteksi dini yang masih rendah. Salah satu metode diagnosis yang umum digunakan adalah imunohistokimia (IHC), yang memungkinkan identifikasi jenis kanker berdasarkan ekspresi protein tertentu seperti Ki-67. Namun, analisis IHC secara manual masih menghadapi berbagai kendala, antara lain waktu pemrosesan yang lama, subjektivitas dalam interpretasi, serta keterbatasan jumlah ahli patologi anatomi. Untuk mengatasi permasalahan tersebut, penelitian ini menerapkan metode deep learning berbasis algoritma You Only Look Once (YOLO) guna meningkatkan akurasi dan efisiensi deteksi sel kanker pada citra IHC kanker payudara. Penelitian ini menggunakan arsitektur YOLOv8 dan YOLOv11 dengan variasi jumlah epoch untuk mengamati pengaruhnya terhadap performa deteksi. Hasil penelitian menunjukkan bahwa model YOLOv8 dan YOLOv11 mampu melakukan pembelajaran secara efektif dengan tren peningkatan performa seiring bertambahnya jumlah epoch. Model YOLOv8s dengan konfigurasi epoch 500 memberikan hasil paling optimal, dengan nilai presisi sebesar 0,75, recall 0,71, F1-score 0,76, dan mAP@50 sebesar 0,76. Model tersebut menunjukkan keseimbangan terbaik antara akurasi deteksi dan kecepatan inferensi. Selain itu, YOLOv8s juga terbukti lebih unggul dibandingkan YOLOv8n, YOLOv11n, dan YOLOv11s dalam hal stabilitas pelatihan dan kemampuan generalisasi pada data uji publik (Kaggle) maupun data uji lokal (RSUA).
=====================================================================================================================================
Breast cancer is the most prevalent type of cancer in Indonesia, with a low rate of early detection. One of the commonly used diagnostic methods is immunohistochemistry (IHC), which enables the identification of cancer types based on the expression of specific proteins such as Ki-67. However, manual analysis of IHC still faces several challenges, including lengthy processing time, subjectivity in interpretation, and a limited number of anatomical pathology experts. To address these issues, this study applies a deep learning approach based on the You Only Look Once (YOLO) algorithm to improve the accuracy and efficiency of cancer cell detection in breast cancer IHC images. The study employs YOLOv8 and YOLOv11 architectures with variations in the number of epochs to observe their impact on detection performance. The results show that both YOLOv8 and YOLOv11 models are capable of effective learning, with a performance improvement trend as the number of epochs increases. The YOLOv8s model with 500 epochs achieved the most optimal results, with a precision of 0.75, recall of 0.71, F1-score of 0.76, and mAP@50 of 0.76. This model demonstrates the best balance between detection accuracy and inference speed. Furthermore, YOLOv8s outperforms YOLOv8n, YOLOv11n, and YOLOv11s in terms of training stability and generalization capability on both public (Kaggle) and local (RSUA) test datasets.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	: YOLOv8, YOLOv11, Deep Learning, Imunohistokimia (IHC) Ki-67, Kanker Payudara, YOLOv8, YOLOv11, Deep Learning, Immunohistochemistry (IHC) Ki-67, Breast Cancer
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QR Microbiology > QR 201.T84 Tumors. Cancer R Medicine > RB Pathology
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Laras Suciningtyas
Date Deposited:	15 Jan 2026 05:39
Last Modified:	15 Jan 2026 05:39
URI:	http://repository.its.ac.id/id/eprint/129652

Actions (login required)

View Item