Muharram, Muhammad Rizki (2025) Penerapan Vision Transformer (ViT) Dalam Diagnosis Penyakit Tuberculosis (TBC) Berdasarkan Citra X-Ray Dada. Undergraduate thesis, Univertias Pembangunan Nasional Veteran Jawa Timur.
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Cover)
21081010275_cover.pdf Download (1MB) |
|
Text (Bab 1)
21081010275_bab 1.pdf Download (251kB) |
|
Text (Bab 2)
21081010275_bab 2.pdf Restricted to Repository staff only until 22 December 2028. Download (562kB) |
|
Text (Bab 3)
21081010275_bab 3.pdf Restricted to Repository staff only until 22 December 2028. Download (730kB) |
|
Text (Bab 4)
21081010275_bab 4.pdf Restricted to Repository staff only until 22 December 2028. Download (1MB) |
|
Text (Bab 5)
21081010275_bab 5.pdf Download (202kB) |
|
Text (Daftar Pustaka)
21081010275_Daftar Pustaka.pdf Download (186kB) |
|
Text (Lampiran)
21081010275_Lampiran.pdf Restricted to Repository staff only until 22 December 2028. Download (262kB) |
Abstract
Tuberculosis (TB) remains one of the most prevalent infectious diseases in Indonesia, requiring rapid, accurate, and consistent diagnostic methods to support early detection. Chest X-ray (CXR) imaging is widely used for initial TB screening; however, its interpretation is often subjective and highly dependent on the radiologist’s experience. This study implements the Vision Transformer (ViT) architecture to automatically classify chest X-ray images for TB detection. The dataset used in this research is the Dataset of Tuberculosis Chest X-rays Images from Mendeley Data, consisting of 3,008 images, including 2,494 TB cases and 514 normal cases. The preprocessing pipeline includes denoising, resizing images to 224×224 pixels, converting grayscale images to RGB, Min-Max normalization, and applying on-the-fly data augmentation. Hyperparameter tuning is performed using Grid Search to determine the optimal combination of batch size, learning rate, and dropout rate for the ViT model. Model performance is evaluated using accuracy, precision, recall, F1-score, along with visualizations such as the confusion matrix and ROC curve. The experimental results indicate that the best-performing ViT configuration, using a batch size of 32, learning rate of 1e-4, and dropout rate of 0.1, achieves exceptionally high performance, with an accuracy of 0.9966, precision of 0.9903, recall of 0.9980, and an F1-score of 0.9941. These findings demonstrate that the Vision Transformer effectively identifies pathological patterns in chest X-ray images and exhibits strong and stable classification capability. Therefore, ViT shows significant potential as a decisionsupport tool for medical image diagnosis, particularly in improving the efficiency, consistency, and accuracy of TB screening.
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > QA Mathematics > QA76.6 Computer Programming T Technology > T Technology (General) |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Informatics | ||||||||||||
| Depositing User: | Muhammad Rizki Muharram | ||||||||||||
| Date Deposited: | 23 Dec 2025 07:40 | ||||||||||||
| Last Modified: | 23 Dec 2025 08:12 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/48487 |
Actions (login required)
 |
View Item |