|
Model name | Purpose | Data type | Result (%) | Strength | Limitation |
|
MV-CNN [54] | Malignant nodule characterization | CT | Acc 92.31 | It is a fast and reliable computer-aided system | A large amount of labeled data is needed for better accuracy |
MP-CNN [333] | Automatic detection of lung cancer | CT | Acc 87.80, spec, 89.10, recall 87.40 | It uses both local and global contextual variables to detect lung cancer | Different image size affects the accuracy |
HSCNN [334] | To predict the malignancy of a pulmonary nodule seen on a computed tomography (CT) scan | CT | Acc 84.40, sens 70.50, spec 88.90, AUC 85.60 | Model interpretability improves with prediction accuracy | No domain specialists can fine-tune it by prioritizing more discriminating features under challenging cases |
NODULEX (CNN features + QIF features) [335] | Differentiate between malignant and benign nodule patterns with accuracy | CT | Acc 94.60, sens, 94.80, spec 94.30 | Excellent accuracy in classifying nodule malignancy | Cross-validated results may be less accurate. Other datasets with significantly differing CT scan picture quality or criteria were not directly fit. |
DENSEBTNET (centercrop operation) [336] | Identifying multiscale features in nodule candidates | CT | Acc 88.31, AUC 93.25 | It has good parameter efficiency and is parameter light. It enhances DenseNet performance and classification accuracy over other approaches. | Its densely connected mechanism causes feature redundancy |
PN-SAMP [337] | Accurately identifying the nodule areas, extracting semantic information from the detected nodules, and predicting the malignancy of the nodules | CT | Acc 97.58 | It can predict the malignancy of lung nodules and offer high-level semantic features and nodule location | Only works on CT images |
Dual-pathway CNN [338] | Predicting the nodule’s malignancy | CT | Acc 86.84 | It performs end-to-end lung nodule diagnostics with high classification accuracy. It can also handle smaller datasets using transfer learning. | A pulmonary nodule cannot be detected automatically |
DeepLung (DUAL-path 3D DCNN+) [71] | Developing a fully automated lung CT cancer detection system | CT | Acc 90.44 | It is smaller and more efficient than residual networks | Lung nodule annotation is not satisfactory |
Ensemble learning of CNNS/multiview knowledge-based collaboration (MV-KBC) [268] | Differentiating between malignant and benign pulmonary nodules | CT | Acc 91.60, AUC 95.70 | It uses an adaptive weighting system learned during error backpropagation to categorize lung nodules, allowing the MV-KBC model to be trained end-to-end | During training, there is a relatively high level of computational complexity |
|