|
| Model name | Purpose | Data type | Result | Strength | Limitation |
|
| RF [303] | Using pretrained model to detect lung cancer accurately | CT | Acc 82.5% | Improves the capacity of lung nodule prediction | Limited dataset and result |
| SVM [300] | Classifying the lung nodules in four lung cancer stages | CT | Acc 84.58% | Predicts small-sized lung nodules, even in low density | The limited dataset affected their results |
| LDA [301] | Classifying cancer using ODNN and LDA | CT | Acc 94.56% | It is quick, easy to use, non-invasive, and inexpensive | Optimal feature selection with multiclassifier was missing |
| RF [304] | Automatic classification of pulmonary peri-fissural nodules (PFNs) | CT | Sens 86.8% | Pretrained CNNs are employed, which makes them faster than training CNNs | All kinds of nodules were not classified |
| SVM [78] | To increase the accurate prediction of lung cancer | CT | Acc 85.7% | Predicts lung cancer from low-resolution data images | The model sometimes fails to predict |
| RF [299] | To detect malignancy of nodules with self-built model NoduleX | CT | Pres 99% | Solid, part-solid, and non-solid nodule categorization is performed automatically | Big nodules were accurately detected |
| RF [305] | Classified the measured solidity or nodules | CT | Acc 95% | Avoids potential errors caused by inaccurate image processing | The description of their work is not described clearly |
| SVM [306] | An improved FP-reduction method is used to detect lung nodules in PET/CT images | CT | Spec 97.2% | Removes around half of the existing FPs | Only small cohort is used |
| Boosting [307] | Classification of nodules with fusion of texture, shape, and deep model-learned data | CT | F1 96.65% | Generates more accurate outcomes than three existing state-of-the-art techniques | The model only detects big nodules |
| Multikernel learning [302] | Distinguishing between the nodule and non-nodule classes with classification | CT | Acc 94.17% | Increases the efficacy of false positive reduction | Dataset name is unclear |
| SVM [308] | Extracting absolute information inherent in raw hand-crafted imaging components | CT | Acc 95.5% | Obtains promising classification outcomes | The reference is limited |
| Decision tree [22] | Using autoencoder with decision tree to detect nodule | CT | Sens 75.01% | Outperforms the state-of-the-art techniques on the overall accuracy measure, even after experimenting with nearly five times the data amount | The results are low |
| SVM [309] | Nodule classification with hybrid features | CT | Acc 99.3% | It extracts the representative image of lung nodule malignancy from chest CT images | The model cannot detect type, position, and size |
| Decision tree [310] | Discovering radiomics to detect lung cancer | CT | Sens 77.52% | Increases the accuracy of lung cancer prediction diagnostics | The reference is limited and results are low |
| Boosting [66] | Identifying nodules from CT scan | CT | AUC 86.42% | Quickly finds the exact positions of latent lung nodule | The references of figure and table are accurately done |
| Multikernel learning [311] | To describe the algorithm for false positive reduction in lung nodule computer-aided detection (CAD) | CT | Jindex 91.39% | Automatically reduces unnecessary feature subsets to get a more discriminative feature set with promising classification performance | All false positive reduction is not done yet |
| Logistic regression [312] | Prediction of the malignancy of lung nodules in CT scans | CT | Sens 94.5% | Additional information based on nodule size has at best a mixed impact on classifier performance | It only takes large nodules |
| DBScan [68] | Detecting nodules with 3D DCNN | CT | Spec 79.67% | It can be expanded into other areas of medical image identification | FP reduction and automated classification are missing |
| Naïve Bayes [243] | A pretrained CNN to extract deep features from lung cancer images and train classifiers to predict all term survivors | CT | Acc 82.5% | The method’s performance is such that adding nodule size information has only a mixed effect on classifier performance | The dataset was too small |
|
