Skripsi

MODIFIKASI U-NET DENGAN SPATIAL PYRAMID POOLING, CONVMIXER, DAN ATTENTION GATE UNTUK SEGMENTASI KANKER PAYUDARA PADA BREAST ULTRASOUND IMAGES

---

Breast cancer is a type of cancer characterized by a lump caused by the growth of abnormal cells in the breast. Early diagnosis of breast cancer is performed by separating normal and abnormal tissue using automatic segmentation. Automatic segmentation can be performed using a Convolutional Neural Network (CNN) architecture, namely U-Net. In the U-Net architecture, feature extraction in the encoder can cause the loss of many important feature information. To address this, the U-Net architecture is modified by applying Spatial Pyramid Pooling (SPP), ConvMixer, and attention gates for breast cancer segmentation on Breast Ultrasound Images (BUSI). SPP is applied after the last encoder layer before the bridge to expand the feature representation. ConvMixer is added after SPP to address feature overlap that arises from merging feature information from multiple spatial levels in SPP. Attention gate is applied in the decoder to filter features and focus the model on capturing relevant information. Image segmentation is performed with two labels, i.e., cancer and background. The breast cancer segmentation results using the proposed architecture achieve an accuracy of 97.7%, indicating the model is excellently in correctly predicting all labels overall. The sensitivity is categorized as good because it shows that 81.57% of cancer areas are successfully detected. The specificity is categorized as very good because it shows that 99.06% of the background can be detected by the model. The IoU shows a fairly good overlap 73.4%, between the predictions and the ground truth. The F1-score is categorized as good because it shows a balance of 84.67% in finding cancer areas and avoiding false detections. These results indicate that modifying the U-Net architecture with SPP, ConvMixer, and attention gates can improve breast cancer segmentation performance on BUSI images. Further research related to classifying cancer types or severity can be carried out based on this study.

---

Availability

Detail Information

Series Title

-

Call Number

T1860262025

Publisher

Indralaya : Prodi Ilmu Matematika, Fakultas Matematika Dan Ilmu Pengetahuan Alam Universitas Sriwijaya., 2025

Collation

xiii, 113 hlm.; ilus.; tab.; 29 cm.

Language

Indonesia

ISBN/ISSN

-

Classification

616.990 7

Content Type

Text

Media Type

unmediated

Carrier Type

other (computer)

Edition

-

Subject(s)

Kanker payudara
Prodi Ilmu Matematika

Specific Detail Info

-

Statement of Responsibility

MI

Other version/related

File Attachment

Comments

---

You must be logged in to post a comment