Skripsi
PENGEMBANGAN ARSITEKTUR 3D U-NET DENGAN RESIDUAL NETWORK DAN INCEPTIONV3 UNTUK SEGMENTASI KANKER PARU PADA CITRA CT SCAN 3D
A lung cancer diagnosis can be performed using 3D CT scan images. Cancerous areas are identified through a segmentation process. This study develops a 3D U-Net with Residual Network (ResNet) and InceptionV3 to segment lung cancer in 3D CT scans. The 3D U-Net architecture was applied because it can exploit volumetric spatial information in 3D CT scan images. ResNet is applied to the encoder section to preserve fine features lost due to downsampling. InceptionV3 is inserted into each ResNet layer to produce a more diverse feature representation. Image segmentation is performed with two labels, namely cancer and background. This study aims to obtain optimal performance from the proposed architecture. The lung cancer segmentation results using the proposed architecture achieved an accuracy of 97%, indicating that the model can accurately classify labels. A sensitivity of 90% and specificity of 99% indicate that the model is more effective at detecting background areas than cancer areas. However, an F1-score of 94% shows that the cancer area detection capability is already very good. An Intersection over Union (IoU) of 88% indicates that the model produces segmentation that matches the ground truth. The results of this study show that modifying the 3D U-Net architecture with ResNet and InceptionV3 can achieve optimal performance in lung cancer segmentation on 3D CT scans.
No other version available