Segmentasi Citra Tulang Alveolar dan Kanal Mandibula pada Citra CBCT Menggunakan Metode YOLOv9

Prambudi, Katarina Inezita (2025) Segmentasi Citra Tulang Alveolar dan Kanal Mandibula pada Citra CBCT Menggunakan Metode YOLOv9. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211148-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Kehilangan gigi berdampak signifikan terhadap kualitas hidup dan implan gigi menjadi solusi umum untuk menggantikan gigi yang hilang. Namun, pemasangan implan harus menghindari kanal mandibula pada tulang alveolar agar tidak menyebabkan kerusakan saraf. Cone Beam Computed Tomography (CBCT) menjadi metode pencitraan yang umum digunakan untuk melihat struktur tulang rahang. Segmentasi tulang alveolar dan kanal mandibula pada citra CBCT sangat penting namun memakan waktu dan bergantung pada keahlian radiolog. Penelitian ini merupakan pengembangan lebih lanjut dari penelitian sebelumnya yang menggunakan You Only Look Once (YOLO) v8 Medium untuk segmentasi citra CBCT. Pengembangan dilakukan karena adanya perbedaan pada dataset, yaitu jumlah data yang meningkat dari 318 menjadi 373 citra CBCT rahang bawah serta adanya ketidakseimbangan distribusi data antara kelas tulang alveolar sebanyak 318 dan kanal mandibula sebanyak 204. Oleh karena itu, pendekatan segmentasi otomatis menggunakan YOLOv9 diterapkan untuk memperoleh performa yang lebih akurat. Dataset dianotasi oleh dokter gigi menggunakan Roboflow dengan pembagian 80% data pelatihan dan 20% data pengujian. Dua arsitektur YOLOv9 (Compact dan Extensive) dilatih dengan variasi hyperparameter berupa jumlah epochs dan batch size. Skenario dilanjutkan dengan preprocessing Histogram Equalization (HE) dan perbandingan terhadap YOLOv8 Medium dari penelitian sebelumnya. Evaluasi dilakukan menggunakan Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Mean Average Precision (mAP50). Model terbaik adalah YOLOv9 Extensive dengan jumlah epochs 75 dan batch size 4 tanpa preprocessing mencapai rata – rata DSC 82,41%, IoU 74%, dan mAP50 93,45%. YOLOv9 terbukti lebih unggul dibanding YOLOv8 Medium yang hanya memperoleh DSC 82,01%. Hasil ini menunjukkan bahwa kombinasi YOLOv9 Extensive tanpa HE efektif untuk segmentasi otomatis tulang alveolar dan kanal mandibula pada citra CBCT.
==================================================================================================================================
Tooth loss significantly affects quality of life and dental implants have become a common solution to replace missing teeth. However, implant placement must avoid the mandibular canal in the alveolar bone to prevent nerve damage. Cone Beam Computed Tomography (CBCT) is a widely used imaging method to visualize jawbone structures. Segmenting the alveolar bone and mandibular canal in CBCT images is crucial but time-consuming and highly dependent on the radiologist’s expertise. This study is a further development of a previous study that used the You Only Look Once (YOLO) v8 Medium model for CBCT image segmentation. The development was motivated by an increased dataset size, from 318 to 373 lower jaw CBCT slices and the presence of class imbalance with 318 alveolar bone versus 204 mandibular canal. Therefore, an automated segmentation approach using YOLOv9 was applied to achieve more accurate performance. The dataset was annotated by professional dentists using Roboflow, with an 80% training and 20% testing data split. Two YOLOv9 architectures (Compact and Extensive) were trained using different hyperparameters, namely the number of epochs and batch sizes. The scenario was extended with Histogram Equalization (HE) preprocessing and a comparison to YOLOv8 Medium from a previous study. Evaluation was conducted using Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and Mean Average Precision (mAP50). The best-performing model was YOLOv9 Extensive with 75 epochs and a batch size of 4 without preprocessing, achieving an average DSC of 82,41%, IoU of 74% and mAP50 of 93,45% and YOLOv9 outperformed YOLOv8 Medium, which only achieved a DSC of 82,01%. These results demonstrate that the YOLOv9 Extensive model without HE is effective for automatic segmentation of alveolar bone and the mandibular canal in CBCT images.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CBCT, Kanal Mandibula, Segmentasi, Tulang Alveolar, YOLOv9, Alveolar Bone, CBCT, Mandibular Canal, Segmentation, YOLOv9
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Katarina Inezita Prambudi
Date Deposited:	07 Jul 2025 04:34
Last Modified:	11 Jul 2025 09:17
URI:	http://repository.its.ac.id/id/eprint/119272

Actions (login required)

View Item