Table 2.
Clinical characteristics of the included studies for the diagnosis of gastrointestinal protruded lesions in wireless capsule endoscopy images using computer-aided diagnosis.
| Study/Year | Nationality of Data | Type of CAD Models | Type of Endoscopic Images | Training Dataset | Type of Test Datasets | Number of Protruded Lesions in Test Dataset | Number of Controls in Test Dataset | TP | FP | FN | TN | Target Conditions |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Li B et al. (2009) [12] | unknown | Feature analysis (texture, color) with MLP | Still cut images | 150 polyp images and 150 normal mucosal images | Internal test | 150 | 150 | 134 | 6 | 16 | 124 | for small bowel polyp diagnosis |
| Hwang S (2011) [13] | unknown | BoW model-SVM | Still cut images | 25 polyp images and 50 normal mucosal images | Internal test | 50 | 100 | 33 | 5 | 17 | 95 | For small bowel polyp diagnosis |
| Karargyris A et al. (2011) [14] | US | Texture analysis with SVM | Still cut images | unclear | Internal test | 10 | 40 | 10 | 13 | 0 | 27 | For small bowel polyp diagnosis |
| Li B et al. (2011) [15] | China | Texture analysis with SVM | Still cut images | 550 tumor images and 550 normal mucosal images | Internal test | 50 | 50 | 45 | 1 | 5 | 49 | for small bowel tumor diagnosis |
| Li B et al. (2011) [16] | China | Texture analysis with an ensemble of kNN, MLP, or SVM | Still cut images | 450 tumor images and 450 normal mucosal images | Internal test | 150 | 150 | 138 | 17 | 12 | 133 | for small bowel tumor diagnosis |
| Barbosa DC et al. (2012) [17] | Portugal | Texture analysis with neural network | Still cut images | 700 tumor images and 2300 normal mucosal images | Internal test | 700 | 2300 | 657 | 159 | 43 | 2141 | for small bowel tumor diagnosis |
| Li B et al. (2012) [18] | China | Texture analysis with SVM | Stil lcut images | 540 tumor images and 540 normal mucosal images | Internal test | 60 | 60 | 51 | 11 | 9 | 49 | for small bowel tumor diagnosis |
| Li B et al. (2012) [19] | China | Texture analysis with SVM | Still cut images | 540 tumor images and 540 normal mucosal images | Internal test | 60 | 60 | 53 | 2 | 7 | 58 | for small bowel tumor diagnosis |
| Constantinescu AF et al. (2015) [20] | Romania | Texture analysis with neural network | Still cut images | unclear | Internal test | 32 | 58 | 30 | 5 | 2 | 53 | for intestinal polyp diagnosis |
| Kundu AK et al. (2020) [21] | from http://www.capsuleendoscopy.org | Linear discriminant analysis with SVM | Still cut images | 30 tumor images and 1617 normal mucosal images | Internal test | 30 | 1617 | 26 | 130 | 4 | 1487 | for small bowel tumor diagnosis |
| Saito H et al. (2020) [22] | Japan | CNN | Still cut images | 30,584 images of protruding lesions | Internal test | 7507 | 10000 | 6810 | 2019 | 697 | 7981 | for protruding lesion diagnosis (small bowel) |
| Yamada A et al. (2020) [23] | Japan | Single Shot MultiBox Detector | Still cut images | 15933 images | Internal test | 1850 | 2934 | 1462 | 380 | 388 | 2554 | for colorectal tumor diagnosis |
CAD, computer-aided diagnosis; TP, true positive; FP, false positive; FN, false negative; TN, true negative; MLP, multilayer perceptron; BoW, Bag-of-Words; SVM, support vector machine; RFE, recursive feature elimination.