Ramadhani, Hanun Masitha (2023) Penentuan Area dan Tingkat Keparahan Nodul pada Citra CT Paru Menggunakan Segmentasi U-Net dan Detaksi Objek YOLO. Masters thesis, Institut Teknologi Sepuluh Nopember.
Text
6025211042-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (2MB) | Request a copy |
Abstract
Tingkat keparahan kanker paru-paru dapat menentukan langkah-langkah pengobatan yang sesuai, sehingga mengurangi resiko kematian akibat kanker paru. Identifikasi tingkat keparahan kanker paru dipantau berdasarkan ukuran nodul kanker dan letak dari nodul untuk menentukan penanganan yang tepat. Kelemahan dari diagnosis kanker paru saat ini adalah mengandalkan ketelitian radiolog sehingga berkemungkinan terjadi kesalahan diagnosis sampai 20%. Oleh karena itu diperlukan sistem identifikasi tingkat keparahan nodul kanker paru berbasis citra CT untuk mengurangi tingkat kesalahan. Namun penelitian sebelumnya masih fokus pada penentuan lokasi nodul kanker paru, belum mengidentifikasi tingkat keparahannya. Penelitian ini mengusulkan sebuah sistem identifikasi untuk menentukan tingkat keparahan dan letak dari nodul kanker paru sehingga diharapkan bisa mengurangi tingkat kesalahan diagnosis.
Penentuan area dan tingkat keparahan berdasarkan proses pelatihan model segmentasi U-Net dan deteksi objek YOLO. Proses pelatihan dilakukan menggunakan data publik Lung Image Database Consortium (LIDC). Metrik yang digunakan untuk menguji model segmentasi U-Net adalah Intersection Over Union (IOU) dan Dice Coefficien. Pada gabungan model U-Net dan MobileNetV2 mendapatkan hasil yang terbaik dengan nilai IoU 0.84 dan Dice 0.81. Sedangkan model deteksi objek menggunakan YOLO untuk menentukan tingkat keparahan juga diukur kinerjanya menggunakan Mean Average Precision (mAP) 0.5. Hasil terbaik didapatkan oleh YOLOv8 dengan nilai mAP 0.5 adalah 0.87. Berdasarkan kedua model tersebut, sistem dapat mendeteksi area dari nodul menggunakan U-Net + MobileNetV2, serta model juga dapat menentukan 4 tingkat keparahan berdasarkan ukuran nodul menggunakan YOLOv8.
=====================================================================================================================================
The severity of lung cancer can determine the appropriate treatment steps, thereby reducing the risk of death from lung cancer. Identification of the severity of lung cancer will be monitored based on the size of the cancer nodule and the location of the nodule to determine the right treatment. The weakness of the current diagnosis of lung cancer is relying on the accuracy of radiologists so that the possibility of a misdiagnosis of up to 20%. Therefore, a CT image-based lung cancer nodule identification system is needed to reduce the error rate. However, previous studies have focused on determining the location of lung cancer nodules, not yet identifying their severity. This study proposes an identification system to determine the severity and location of lung cancer nodules so that it is expected to reduce the misdiagnosis rate.
Area and severity determination based on training process of U-Net segmentation model and YOLO object detection. The training process is carried out using the Lung Image Database Consortium (LIDC) public data. The metrics used to test the U-Net segmentation model are Intersection Over Union (IOU) and Dice Coefficient. The combined U-Net and MobileNetV2 models got the best results with IoU values of 0.84 and Dice 0.81. While the object detection model uses YOLO to determine the level of severity, its performance is also measured using a Mean Average Precision (mAP) 0.5. The best result was obtained by YOLOv8 with a mAP value of 0.5 which was 0.87. Based on these two models, the system can detect areas of nodules using U-Net + MobileNetV2, and the model can also determine 4 severity levels based on nodule size using YOLOv8.
Item Type: | Thesis (Masters) |
---|---|
Uncontrolled Keywords: | Kanker paru, Segmentasi, U-Net + MobileNetv2, Deteksi Objek, YOLOv8, Lung Cancer, Segmentasi, U-Net + MobileNetv2, Object Detection, YOLOv8 |
Subjects: | R Medicine > R Medicine (General) > R858 Deep Learning R Medicine > RC Internal medicine > RC78 Diagnosis, Radioscopic--Examinations, questions, etc. |
Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis |
Depositing User: | Hanun Masitha Ramadhani |
Date Deposited: | 22 Jul 2023 13:23 |
Last Modified: | 22 Jul 2023 13:23 |
URI: | http://repository.its.ac.id/id/eprint/98882 |
Actions (login required)
View Item |