Andrianto, Andrianto (2025) Menggunakan Pembelajaran Mendalam untuk Menghasilkan Gambar Ultrasonografi B-Mode Sintetis. (Using Deep Learning to Generate Synthetic B-Mode Ultrasonography Images). Undergraduate thesis, Universitas 17 Agustus 1945 Surabaya.
|
Text (ABSTRAK)
abstrak.pdf Download (11MB) | Preview |
|
![[img]](http://repository.untag-sby.ac.id/style/images/fileicons/text.png) |
Text (BAB I)
TUGAS_AKHIR-bab 1- hal - 27-30.pdf Restricted to Repository staff only Download (286kB) | Request a copy |
|
|
Text (BAB II)
TUGAS_AKHIR-bab 2- hal - 31-54.pdf Restricted to Repository staff only Download (986kB) | Request a copy |
|
|
Text (BAB III)
TUGAS_AKHIR-bab 3- hal - 55-86-1-8.pdf Restricted to Repository staff only Download (319kB) | Request a copy |
|
|
Text (BAB IV)
TUGAS_AKHIR-bab 3- hal - 55-86-9-32.pdf Restricted to Repository staff only Download (820kB) | Request a copy |
|
|
Text (BAB V)
TUGAS_AKHIR-87-90.pdf Restricted to Repository staff only Download (285kB) | Request a copy |
|
|
Text (JURNAL)
jurnal.pdf Download (730kB) | Preview |
|
|
Text (JURNAL TURNITIN)
turnitin_1462100261.pdf Download (535kB) | Preview |
Abstract
B-mode ultrasound imaging is one of the essential medical imaging modalities used to diagnose various internal conditions of the body, such as soft tissue and muscle structures. However, limitations of portable ultrasound devices often result in low-resolution images with high noise levels, which can reduce the accuracy of medical diagnoses. This study employs a deep learning approach based on Generative Adversarial Networks (GAN), specifically using the CycleGAN architecture, to generate realistic synthetic B-mode ultrasound images without relying on paired data. The dataset used consists of two-dimensional ultrasound images of fetal heads in a transverse plane during the second trimester, obtained from the public Zenodo repository. Prior to model training, unpaired data from two domains were used. To improve data quality, preprocessing steps such as normalization and histogram equalization were applied. Additionally, the model architecture includes two generators and two discriminators, with a histogram-based discriminator added to enhance the similarity of pixel distributions. The results were comprehensively evaluated using both quantitative and visual metrics, including Fréchet Inception Distance (FID), Histogram Intersection, Entropy, and Skewness. Training results indicate that the model was able to generate synthetic images with visual and statistical qualities close to the original images, with FID significantly decreasing to 1208.49 and a histogram intersection of 0.8113 at epoch 77. The differences in entropy and skewness between real and generated images also became smaller, indicating the model's ability to preserve important anatomical structures. This study demonstrates that CycleGAN can be effectively used to generate high-quality B-mode ultrasound images from portable devices without relying on paired data. These findings make a significant contribution to the field of medical data augmentation, AI-based diagnostic model training, and improving diagnostic efficiency in clinical environments with limited devices and data availability
| Item Type: | Thesis (Undergraduate) |
|---|---|
| Uncontrolled Keywords: | CycleGAN, GAN, Citra Ultrasonografi Sintetis, Pembelajaran Mendalam, Histogram Intersection. |
| Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science R Medicine > R Medicine (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Fakultas Teknik > Program Studi Teknik Informatika |
| Depositing User: | 1462100261 Andrianto Andrianto |
| Date Deposited: | 27 Aug 2026 06:14 |
| Last Modified: | 27 Aug 2026 06:14 |
| URI: | http://repository.untag-sby.ac.id/id/eprint/45858 |
Actions (login required)
 |
View Item |
