Table 3.
Most relevant studies on video capsule AI application in CD.
| Author (Year) | Study Design | Population | Aim | Results |
|---|---|---|---|---|
| Girgis et al. (2010) | Retrospective cohort study | 47 videos from 29 CD, 17 control, 1 celiac patient | To test a system able to detect inflammation among the thousands of images acquired by the WCE | Total accuracy, specificity, and sensitivity of 87%, 93%, and 80%, respectively |
| Kumar et al. (2012) | Retrospective cohort study | 47 videos, 30 of which contained CD lesions |
To test a supervised classification for CD lesions and for quantitative assessment of lesion severity | Good precision (>90% for lesion detection) and recall (>90%) for lesions of varying severity |
| Charisis et al. (2016) | Retrospective cohort study | 800-image database from 13 CD patients | To test HAF-DLac approach for automated lesion detection | Accuracy, sensitivity, specificity, and precision of 93.8%, 95.2%, 92.4%, and 92.6%, respectively |
| Klang et al. (2020) | Retrospective cohort study | 17,640 CE images from 49 CD patients | To test a CNN in classifying images into either normal mucosa or mucosal ulcers | AUC of 0.99 and accuracy ranging from 95.4% to 96.7% |
| Klang et al. (2021) | Retrospective cohort study | 27,892 CE images | To test a DLN for detecting CE images of strictures | For classification of strictures vs. nonstrictures, average accuracy of 93.5% (±6.7%) |
| Barash et al. (2021) | Retrospective cohort study | 17,640 CE images from 49 CD patients | To test a CNN in automatically grading images of ulcers and compare the resulting algorithm with a consensus reading | Algorithm accuracy of 0.91 for grade 1 vs. grade 3 ulcers, of 0.78 for grade 2 vs. grade 3, and of 0.62 for grade 1 vs. grade 2 |
| Majtner et al. (2021) | Retrospective cohort study | 7744 images from 38 CD patients (small bowel 4972, colon 2772) | To test the ability of a DL framework to detect lesions with panenteric capsule endoscopy | Diagnostic accuracy of 98.5% for small bowel and 98.1% for colon |
| Ferreira JPS et al. (2021) | Retrospective cohort study | 8085 images | To develop and validate a CNN for ulcer and erosion detection using panenteric capsule endoscopy images | Model sensitivity, specificity, precision, and accuracy of 90.0%, 96.0%, 97.1%, and 92.4%, respectively |
Abbreviations: AUC: area under the curve; CD: Crohn’s Disease; CE: capsule endoscopy; CNN: convolutional neural network; DL: deep learning; DLac: differential lacunarity; DLN: deep learning network; HAF: hybrid adaptive filtering; WCE: wireless capsule endoscopy.