Skripsi

PERBAIKAN KUALITAS CITRA JANTUNG MENGGUNAKAN METODE DEEP LEARNING

---

In infant heart ultrasound images, problems often appear such as low brightness, noise and blur. To overcome this problem, an improvement process is needed using deep learning techniques such as MIRNet, Autoencoder, EDR-CNNs, and LLCNN. The best method is MIRNet, which produces PSNR values of 36.37 dB, MSE of 16.69, and SSIM of 92.86. In addition, the results obtained were tested through classification into four classes (ASD, AVSD, NORMAL, VSD) and two classes (Normal and Abnormal). The best results from classification using unseen MIRNet and transfer learning with VGG19 show improvement after the enhancement process. For two class classification, the validation accuracy before enhancement was 95.94%. After using model enhancement, accuracy increased to 98.37%. Unseen accuracy before enhancement was 71.27%, respectively, and increased to 76.59% after enhancement. For four-class classification, the validation accuracy before enhancement was 98.91% for ASD, 100% for AVSD, 97.83% for NORMAL, and 98.91% for VSD. After enhancement, accuracy increased to 100% for ASD and AVSD, and 99.45% for NORMAL and VSD. Unseen accuracy before enhancement was 75.00% for ASD, 87.23% for AVSD, 73.93% for NORMAL, and 82.97% for VSD. After enhancement, the accuracy for ASD, AVSD, NORMAL, and VSD were 77.65%, 87.76%, 77.65%, and 84.57%, respectively.

---

Availability

Detail Information

Series Title

-

Call Number

T1561722024

Publisher

Indralaya : Prodi Sistem Komputer, Fakultas Ilmu Komputer Universitas Sriwijaya., 2024

Collation

xiv, 62 hlm.; ilus.; tab, 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

006.370 7

Content Type

Text

Media Type

unmediated

Carrier Type

other (computer)

Edition

-

Subject(s)

Prodi Sistem Komputer
Visi Komputer

Specific Detail Info

-

Statement of Responsibility

SEW

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment