Skripsi

PENGEMBANGAN MODEL SEGMENTASI LESI CITRA PRA-KANKER SERVIKS MENGGUNAKAN YOLO DAN SAM 2 BERBASIS KONTUR

---

Cervical cancer is one of the leading causes of morbidity and mortality among women, making early detection of precancerous lesions essential. However, lesion segmentation in cervical images still faces several challenges, including unclear object boundaries, illumination variations, imaging artifacts, and class imbalance between lesion and background, which reduce the performance of deep learning-based models. Therefore, this study aims to develop a lesion segmentation model by integrating You Only Look Once (YOLO) and the Segment Anything Model (SAM) using a contour-based prompting strategy within a two-stage framework, where YOLOv11-seg generates the initial segmentation and SAM 2 performs refinement to improve object boundaries. This study also evaluates contour-based SAM 3 and implements a rule-based color fallback mechanism to handle segmentation failures in the columnar area. The dataset consists of 881 cervical precancer images obtained from IVA examinations with three main classes: lesion, columnar area, and cervical area. Model performance was evaluated using Intersection over Union (IoU) and Dice Coefficient metrics. The results show that contour-based prompting improves segmentation quality compared to bounding box and centroid-based methods and is more effective in capturing complex and irregular lesion boundaries. The integration of YOLO and SAM 2 also provides more stable segmentation performance compared to YOLO-only and SAM 3-based refinement approaches for lesion segmentation. In addition, the color fallback mechanism helps maintain columnar area segmentation when the primary model prediction fails. Therefore, the proposed model has the potential to support more precise, adaptive, and efficient AI-based cervical cancer early detection systems.

---

Availability

Detail Information

Series Title

-

Call Number

T2000472026

Publisher

Indralaya : Prodi Sistem Komputer, Fakultas Ilmu Komputer Universitas Sriwijaya., 2026

Collation

xviii, 181 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

618.140 7

Content Type

Text

Media Type

-

Carrier Type

-

Edition

-

Subject(s)

Prodi Sistem Komputer
Penyakit Kanker Serviks

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment