Table 2.

The table lists the characteristics of different studies aiming at lung nodules screening and classification.

Authors	Country	Imaging Modality	Patient Number	Study Model	AI System	Validation	Main Theme	Strengths	Weakness
Nasrullah et al. [48]	China	LDCT	LIDC-IDRI dataset	Retrospective	Two deep 3D customized mixed link network architectures for lung nodule detection and classification	LIDC-IDRI and LUNA 16 dataset	Lung nodule detection and classification	The system achieved promising results in the form of sensitivity (94%) and specificity (91%)	Validation only in pre-clinical settings
Kopelowitz et al. [49]	U.K.	CT	LUNA 16 dataset	Retrospective	Modified MaskRCNN to handle 3D images	LUNA 16 dataset	Lung nodule detection and segmentation	All-in-one system for detection and segmentation	Validation only on the LUNA 16 dataset
Ding et al. [50]	China	CT	LUNA16 dataset	retrospective	Faster R-CNN for detection and three-dimensional DCNN for the subsequent false positive reduction	LUNA16 dataset	Lung nodule detection	Good detection performance on nodule detection ranking the 1st place of Nodule Detection Track (NDET) in 2017	Needs validation on bigger datasets
Khosravan et al. [51]	U.S.A.	CT	LUNA16 dataset	retrospective	3D densely connected CNN	LUNA16 dataset	Lung nodule detection	single-shot single-scale fast lung nodule detection algorithm without the need for additional FP removal	Validation only on the LUNA 16 dataset
Tran et al. [55]	Vietnam, France	CT	LUNA16 dataset	retrospective	15-layer 2D deep CNN architecture (LdcNet)	LUNA16 dataset	automatic feature extraction and classification of pulmonary candidates as nodule or non-nodule	High-quality classifier with an accuracy of 97.2%, sensitivity of 96.0%, and specificity of 97.3%.	Only validated in one preclinical dataset
Wu et al. [56]	China, U.S.A., Australia, U.K., Germany	CT	LIDC-IDRI dataset	Retrospective	50-layer deep residual network	LIDC-IDRI dataset	Lung nodule classification	The lung nodule image can be used as the input data of the network directly, avoiding complicated feature extraction and selection.	Long training time is needed when dealing with a large number of lung CT images
Mastouri et al. [57]	Tunisia	CT	LUNA16 dataset	Retrospective	Three bilinear-CNN followed by a linear SVM classifier	LUNA16 dataset	Lung nodule classification	The system was validated on the LUNA16 dataset and compared to the outcomes of conventional CNN-based architectures showing promising and satisfying results	The bilinear pooling requires massive calculation and storage costs, making this algorithm impractical
Al-Shabi et al. [58]	Malaysia, Singapore, U.S.A.	CT	LIDC-IDRI dataset	Retrospective	Gated Dilated(GD) network	LIDC-IDRI dataset	Classification of pulmonary nodules as benign or malignant	Better discrimination whether benign or malignant for mid-sized nodules	Requires an object detector model to identify the nodule locations before classifying them as benign/malignant
Liu et al. [59]	China, U.S.A.	CT	LIDC-IDRI dataset	Retrospective	multi-model ensemble learning architecture based on 3D convolutional neural network (MMEL-3DCNN)	LIDC-IDRI dataset	Benign/malignant lung nodule classification	Image enhancement on the input data to improve the contrast of lung nodules with low contrast to surrounding tissues	Validation only in pre-clinical setting