We proposed a end-to-end architecture based on 3D U-Net convolutional neural network that performs 3D feature extraction and classification of lung nodules in CT scans.
We extended the previous 2D U-Net architecture to 3D so that 3D features of lung nodules in CT scans can be effectively used for nodule detection.
In addition, we defined a suitable loss function to train 3D U-Net model with the mask data obtained from LIDC dataset to improve the performance of lung nodule detection.
The experiment results on LIDC dataset showed that the proposed method can detect nodules of 3 mm to 30 mm and outperform the previous methods in sensitivity of lung nodule detection.
The proposed lung nodule detection method can be used effectively to support early diagnosis of lung cancer in clinical practice, as it can detect the lung nodules as small as 3 mm without missing them.
This method was published in "International Journal of Advanced Networking and Applications" under the title of "A New Method of Lung Nodule Detection in CT Scans 3D U-Net Convolutional Neural Network" (https://doi.org/10.35444/IJANA.2024.15501).
