Segmentation of Lung Nodules on CT Images Using a Nested Three-Dimensional Fully Connected Convolutional Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate lung nodule segmentation is crucial for computer-aided diagnosis of lung cancer.
• Conventional methods struggle with nodules obscured by the chest wall or ground-glass opacities.

Purpose of the Study:

• To develop a robust and accurate 3D segmentation method for lung nodule regions using deep learning.
• To improve the performance of lung nodule segmentation in computed tomography (CT) images.

Main Methods:

• A nested 3D fully connected convolutional network with residual units was proposed.
• A novel loss function was designed for improved segmentation accuracy.
• The method was evaluated on 332 lung adenocarcinoma nodules from 332 patients.

Main Results:

• The proposed method achieved a Dice similarity coefficient (DS) of 0.845 ± 0.008 and an intersection over union (IoU) of 0.738 ± 0.011.
• It demonstrated superior performance compared to 3D U-Net (DS: 0.822 ± 0.009, IoU: 0.711 ± 0.011) and 3D SegNet (DS: 0.786 ± 0.011, IoU: 0.660 ± 0.012).
• The method significantly outperformed conventional watershed (DS: 0.628 ± 0.027, IoU: 0.494 ± 0.025) and graph cut (DS: 0.566 ± 0.025, IoU: 0.414 ± 0.021) methods.

Conclusions:

• The proposed deep learning method provides accurate and robust 3D segmentation of lung nodules.
• This technique can aid radiologists in diagnosing lung nodules, including lung adenocarcinoma, on CT images.