Automated Segmentation of Lung Tomography Images Using the U-Net Model: A Pilot Study

The segmentation of lung computerized tomography (CT) images using the U-Net model, a type of convolutional neural network, was aimed in this study. 30 lung CT images obtained from the Department of Radiation Oncology at Ankara University Faculty of Medicine were utilized. Twenty of these images were used during the training phase of the model, while ten were used during the testing phase. The CT images, with dimensions of $512 \times 512$ pixels and containing 256 slices, were converted from DICOM format to NIfTI format. For the training phase, the 20 images were augmented to 100 images using various image augmentation techniques. During the testing phase, the lung structures of the 10 images were segmented using the Slicer program. Using the U-Net model, the total 2560 slices in the test images were compared with the Slicer segmentations, and the average dice index was obtained as 96.44±2.62. The results demonstrate that the U-Net model provides high accuracy in the segmentation of lung CT images.