Color masking improves classification of celiac disease in videocapsule endoscopy images

Artificial Intelligence-Assisted Capsule Endoscopy Versus Conventional Capsule Endoscopy for Detection of Small Bowel Lesions: A Systematic Review and Meta-Analysis

BACKGROUND

Capsule endoscopy (CE) is a valuable tool used in the diagnosis of small intestinal lesions. The study aims to systematically review the literature and provide a meta-analysis of the diagnostic accuracy, specificity, sensitivity, and negative and positive predictive values of AI-assisted CE in the diagnosis of small bowel lesions in comparison to CE.

METHODS

Literature searches were performed through PubMed, SCOPUS, and EMBASE to identify studies eligible for inclusion. All publications up to 24 November 2024 were included. Original articles (including observational studies and randomized control trials), systematic reviews, meta-analyses, and case series reporting outcomes on AI-assisted CE in the diagnosis of small bowel lesions were included. The extracted data were pooled, and a meta-analysis was performed for the appropriate variables, considering the clinical and methodological heterogeneity among the included studies. Comprehensive Meta-Analysis v4.0 (Biostat Inc.) was used for the analysis of the data.

RESULTS

A total of 14 studies were included in the present study. The mean age of participants across the studies was 54.3 years (SD 17.7), with 55.4% men and 44.6% women. The pooled accuracy for conventional CE was 0.966 (95% CI: 0.925-0.988), whereas for AI-assisted CE, it was 0.9185 (95% CI: 0.9138-0.9233). Conventional CE exhibited a pooled sensitivity of 0.860 (95% CI: 0.786-0.934) compared with AI-assisted CE at 0.9239 (95% CI: 0.8648-0.9870). The positive predictive value for conventional CE was 0.982 (95% CI: 0.976-0.987), whereas AI-assisted CE had a PPV of 0.8928 (95% CI: 0.7554-0.999). The pooled specificity for conventional CE was 0.998 (95% CI: 0.996-0.999) compared with 0.5367 (95% CI: 0.5244-0.5492) for AI-assisted CE. Negative predictive values were higher in AI-assisted CE at 0.9425 (95% CI: 0.9389-0.9462) versus 0.760 (95% CI: 0.577-0.943) for conventional CE.

CONCLUSION

AI-assisted CE displays superior diagnostic accuracy, sensitivity, and positive predictive values albeit the lower pooled specificity in comparison with conventional CE. Its use would ensure accurate detection of small bowel lesions and further enhance their management.

Computer aided villi morphometric quantification in video-capsule enteroscopy: A newly developed software to quantify small bowel atrophy

BACKGROUND AND AIMS

Small bowel capsule endoscopy (SBCE) has an established role in patients with non-responsive celiac disease (CeD). A non-invasive method to quantify small bowel atrophy is still lacking.

METHODS

We analysed SBCE frames from CeD patients from 2018 to 2020. Histology was the reference standard, with atrophy defined as Marsh-Oberhuber score ≥ 3a. Three regions of interest (ROI) were blindly selected from each frame by an expert gastroenterologist and analysed using a National Institute of Health J image-processing software into a numerical scale. A 3D surface plot macro identified intestinal villi density through isolines plots.

RESULTS

We acquired 306 ROIs from 57 frames with macroscopic atrophy and 45 with normal mucosa. Frames were classified as atrophic (n = 63) or non-atrophic (n = 39) per Marsh-Oberhuber classification. Median density score significantly differed between atrophic and non-atrophic frames (p < 0.001). The morphometric analysis showed a sensitivity of 77 % and a specificity of 79 % in discriminating between atrophic or non-atrophic mucosa with a 14.10 cut-off (Youden Index) and an overall AUC of 0.805 (CI 95 % 0.712-0.897).

CONCLUSIONS

Our newly developed SBCE software can effectively quantify villous atrophy. Further studies are needed to validate its applicability in an external cohort.

Celiac Disease Deep Learning Image Classification Using Convolutional Neural Networks

Celiac disease (CD) is a gluten-sensitive immune-mediated enteropathy. This proof-of-concept study used a convolutional neural network (CNN) to classify hematoxylin and eosin (H&E) CD histological images, normal small intestine control, and non-specified duodenal inflammation (7294, 11,642, and 5966 images, respectively). The trained network classified CD with high performance (accuracy 99.7%, precision 99.6%, recall 99.3%, F1-score 99.5%, and specificity 99.8%). Interestingly, when the same network (already trained for the 3 class images), analyzed duodenal adenocarcinoma (3723 images), the new images were classified as duodenal inflammation in 63.65%, small intestine control in 34.73%, and CD in 1.61% of the cases; and when the network was retrained using the 4 histological subtypes, the performance was above 99% for CD and 97% for adenocarcinoma. Finally, the model added 13,043 images of Crohn's disease to include other inflammatory bowel diseases; a comparison between different CNN architectures was performed, and the gradient-weighted class activation mapping (Grad-CAM) technique was used to understand why the deep learning network made its classification decisions. In conclusion, the CNN-based deep neural system classified 5 diagnoses with high performance. Narrow artificial intelligence (AI) is designed to perform tasks that typically require human intelligence, but it operates within limited constraints and is task-specific.

Celiac disease diagnosis from videocapsule endoscopy images with residual learning and deep feature extraction

BACKGROUND AND OBJECTIVE

Videocapsule endoscopy (VCE) is a relatively new technique for evaluating the presence of villous atrophy in celiac disease patients. The diagnostic analysis of video frames is currently time-consuming and tedious. Recently, computer-aided diagnosis (CAD) systems have become an attractive research area for diagnosing celiac disease. However, the images captured from VCE are susceptible to alterations in light illumination, rotation direction, and intestinal secretions. Moreover, textural features of the mucosal villi obtained by VCE are difficult to characterize and extract. This work aims to find a novel deep learning feature learning module to assist in the diagnosis of celiac disease.

METHODS

In this manuscript, we propose a novel deep learning recalibration module which shows significant gain in diagnosing celiac disease. In this recalibration module, the block-wise recalibration component is newly employed to capture the most salient feature in the local channel feature map. This learning module was embedded into ResNet50, Inception-v3 to diagnose celiac disease using a 10-time 10-fold cross-validation based upon analysis of VCE images. In addition, we employed model weights to extract feature points from training and test samples before the last fully connected layer, and then input to a support vector machine (SVM), k-nearest neighbor (KNN), and linear discriminant analysis (LDA) for differentiating celiac disease images from heathy controls.

RESULTS

Overall, the accuracy, sensitivity and specificity of the 10-time 10-fold cross-validation were 95.94%, 97.20% and 95.63%, respectively.

CONCLUSIONS

A novel deep learning recalibration module, with global response and local salient factors is proposed, and it has a high potential for utilizing deep learning networks to diagnose celiac disease using VCE images.