Table 2.
Analysis of Previous work of the researchers to detect and classify gastric diseases.
| Author’s Name | ||||
|---|---|---|---|---|
| Li et al. (2018) (12) | 560 gastric slices, 140 normal slices | Gastric Net | Average classification accuracy = 97.93% | Limited dataset |
| Nadeem et al. (2018) (10) | MediaEval 8000 images | VGG-19, logistic regression | Accuracy= 83% | Low accuracy |
| Hirasawa et al. (2018) (13) | 13584 endoscopic images | CNN | Sensitivity = 92.2% Positive predicted value = 30.6% |
The model was trained with the high quality images which means it won’t work for post-biopsy bleeding, images and less insufflation of air. |
| Song et al. (2020) (14) | PLAGH dataset | DeepLabV3 | Average specificity = 80.6% | The cost of computation was high |
| Yuan & Meng (2017) (20) | WCE image dataset | stacked sparse autoencoder | Accuracy= 98% | Needs improvement in classification accuracy |
| Mortezagholi et al. (2019) (15) | Samples of 405 patients | KNN, Naïve Bayes, SVM | Accuracy = 90.8% F1 score = 91.99% |
class imbalance |
| Aslam et al. (2020) (16) | Data collected from 220 samples of cancerous and non-cancerous stomach | Support vector machine, linear kernel | Accuracy = 97.18% Specificity = 97.44% F1 score = 91.99% |
SVM failed to predict 14 instances |
| Ueyama et al. (2021) (17) | Dataset of 5574 magnifying narrow brand imaging | Deep learning, CAD system | Accuracy = 98.77% Specificity = 100% Sensitivity = 98% |
Sometime it was difficult for the model to distinguish from gastritis. |
| Zhou et al. (2014) (19) | 359 frames of video capsule endoscopy | Support vector machine classifier | Accuracy= 90.77% | Some algorithms must be incorporated for the robustness of the approach |
| Liu et al. (2019) (18) | 557 patients of gastric cancer | Support vector machine, autoencoder, logistic regression | Accuracy = 89% Specificity = 79% Sensitivity = 78% F1 Score = 95% |
Small dataset |
| Asperti & Mastronardo (2017) (22) | Kvasir dataset | Inception | Accuracy=91.55% Precision = 91.5% |
The system was required to diagnose other GI tract-based disorders. |
| Liu et al. (2018) (21) | Dataset of Gastric pathology images | Artificial neural network | F-score = 0.96 | This approach needs further improvement of classification accuracy |
| Sun et al. (2020) (24) | Dataset of annotated gastricscopic images | Deep neural network | Accuracy = 96.7% Recall = 94.9% F1score = 94.7% |
Overfitting |