|
Model name | Purpose | Data type | Result (%) | Strength | Limitation |
|
DBN with RBM [317] | To detect nodules with deep networks | CT | Acc 92.83 | No relative location information is ignored to extract features that express the original image better | The references were very limited with less info of method |
DRL [318] | Detecting lung cancer with several potential deep reinforcement learning models | CT | Acc 80 | Got promising results in tumor localization | The result of their work is not fully cleared |
DRN [319] | Detecting lung cancer in FDG-PET imaging under ultra-low-dose PET scans | PET | Acc 97.1 | Lung cancer detection is automated even at low effective radiation doses | The outcome is insufficient |
DBN with RBM [320] | Testing the feasibility of using DL algorithms for lung cancer diagnosis | CT | Acc 79.40 | It has shown very promising results | Accuracy was slightly less than CNN model |
Deep denoising autoencoder [321] | A combination of deep-learned representations was employed to create a lengthy feature vector, which was then used to train the classification of nodules | CT | Acc 95.5 | Increased the ability to differentiate between malignant and benign nodules, with a significant improvement in sensitivity | The dataset was not a benchmarked dataset |
DRN [322] | Training model first and applying 3D ConvNet to detect lung nodule with hybrid loss learning | CT | Acc 86.7 | It detects pulmonary nodules from low-dose CT scans | Detects small nodules and cannot classify malignant or benign nodules |
DBN with RBM [23] | Comparing DL and CNN model on lung nodule detection | CT | Sens 73.4 | It solves the longstanding challenge of classifying lung nodules as malignant or benign without computing morphological or textural data | The classification was very limited |
DRN [323] | Identification of lung nodules from CT scans is efficient for lung cancer diagnosis, and false positive reduction is important, so it was the aim | CT | Acc 98 | It is reliable and detects well. It may also be easily extended to detect 3D objects. | Figures and table are not referred clearly |
DRL [77] | Developing and validating a reinforcement learning model for early identification of lung nodules in CT images | CT | Acc 99.1 | Eliminated the major issue of false positives in CT lung nodule screening, saving unwanted tests and expenditures | Only the big nodules were detected |
Deep denoising autoencoder [324] | A spherical harmonic expansion is used as it has ability to approximate the surfaces of tough shapes of the detected lung nodules | CT | Acc 96 | It can show small or big lung nodule spatial inhomogeneities | Classification of nodule as malignant or benign was not done |
Multilayer perceptron model [325] | To analyze the performance of several ML methods for detecting lung cancer | CT | Acc 88.55 | The presented image preprocessing method detects cancerous bulk | The layers of the model were not discussed briefly |
Deep stacked autoencoder [326] | The main purpose is to train a 3D CNN with data and convert it into a 3D fully convolutional network (FCN) that can generate the score map | CT | Sens 80 | It can generate the score map for the whole volume in a single pass | The results were not compared with other models |
Deep sparse autoencoder [327] | Analyzing the nodules of CT data and helping the experts to be more the accurate with proposed analysis tool | CT | Acc 99.57 | Improving the display of actual medical CT data may automatically extract pulmonary nodule features | The information of dataset is missing |
GAN [328] | Building a 3D U-Net and CNN to segment and identify nodule and assist the radiologists understand CT images | CT | Acc 95.4 | Malignant nodule detection is precise and effective | Detects large nodule more accurately than the small nodules |
Deep stacked autoencoder [260] | To get an accurate diagnosis of the detected lung nodules | CT | Acc 92.20 | It classified nodules using higher-order MGRF and geometric criteria | They did not mention any reshape or resize techniques |
|