Computer-aided classification of colorectal segments during colonoscopy: a deep learning approach based on images of a magnetic endoscopic positioning device

OBJECTIVE

Assessment of the anatomical colorectal segment of polyps during colonoscopy is important for treatment and follow-up strategies, but is largely operator dependent. This feasibility study aimed to assess whether, using images of a magnetic endoscope imaging (MEI) positioning device, a deep learning approach can be useful to objectively divide the colorectum into anatomical segments.

METHODS

Models based on the VGG-16 based convolutional neural network architecture were developed to classify the colorectum into anatomical segments. These models were pre-trained on ImageNet data and further trained using prospectively collected data of the POLAR study in which endoscopists were using MEI (3930 still images and 90,151 video frames). Five-fold cross validation with multiple runs was used to evaluate the overall diagnostic accuracies of the models for colorectal segment classification (divided into a 5-class and 2-class colorectal segment division). The colorectal segment assignment by endoscopists was used as the reference standard.

RESULTS

For the 5-class colorectal segment division, the best performing model correctly classified the colorectal segment in 753 of the 1196 polyps, corresponding to an overall accuracy of 63%, sensitivity of 63%, specificity of 89% and kappa of 0.47. For the 2-class colorectal segment division, 1112 of the 1196 polyps were correctly classified, corresponding to an accuracy of 93%, sensitivity of 93%, specificity of 90% and kappa of 0.82.

CONCLUSION

The diagnostic performance of a deep learning approach for colorectal segment classification based on images of a MEI device is yet suboptimal (clinicaltrials.gov: NCT03822390).

Reliability of Measuring Ileo-Colonic Disease Activity in Crohn's Disease by Magnetic Resonance Enterography

BACKGROUND

Magnetic resonance enterography is increasingly utilized for assessment of luminal Crohn's disease activity. The Magnetic Resonance Index of Activity and the London Index are the most commonly used outcome measures in clinical trials. We assessed the reliability of these indices and several additional items.

METHODS

A consensus process clarified scoring conventions and identified additional items based on face validity. Four experienced radiologists evaluated 50 images in triplicate, in random order, at least 1 month apart, using a central image management system. Intra- and interrater reliability were assessed by calculating and comparing intraclass correlation coefficients.

RESULTS

Intrarater intraclass correlation coefficients (95% confidence intervals) for the Magnetic Resonance Index of Activity, London, and London "extended" indices and a visual analogue scale were 0.89 (0.84 to 0.91), 0.87 (0.83 to 0.90), 0.89 (0.85 to 0.92), and 0.86 (0.81 to 0.90). Corresponding interrater intraclass correlation coefficients were 0.71 (0.61 to 0.77), 0.67 (0.55 to 0.75), 0.70 (0.61 to 0.76), and 0.71 (0.62 to 0.77). Reliability for each index was greatest in the terminal ileum and poorest in the rectum. All 3 indices were highly correlated with the visual analogue scale; 0.79 (0.71 to 0.85), 0.78 (0.71 to 0.84), and 0.79 (0.72 to 0.85) for the Magnetic Resonance Index of Activity, London, and the London "extended" indices, respectively.

CONCLUSIONS

"Substantial" interrater reliability was observed for all 3 indices. Future studies should assess responsiveness to treatment in order to confirm their utility as evaluative indices in clinical trials and clinical practice.

Development of the Lémann index to assess digestive tract damage in patients with Crohn's disease

BACKGROUND & AIMS

There is a need for a scoring system that provides a comprehensive assessment of structural bowel damage, including stricturing lesions, penetrating lesions, and surgical resection, for measuring disease progression. We developed the Lémann Index and assessed its ability to measure cumulative structural bowel damage in patients with Crohn's disease (CD).

METHODS

We performed a prospective, multicenter, international, cross-sectional study of patients with CD evaluated at 24 centers in 15 countries. Inclusions were stratified based on CD location and duration. All patients underwent clinical examination and abdominal magnetic resonance imaging analyses. Upper endoscopy, colonoscopy, and pelvic magnetic resonance imaging analyses were performed according to suspected disease locations. The digestive tract was divided into 4 organs and subsequently into segments. For each segment, investigators collected information on previous operations, predefined strictures, and/or penetrating lesions of maximal severity (grades 1-3), and then provided damage evaluations ranging from 0.0 (no lesion) to 10.0 (complete resection). Overall level of organ damage was calculated from the average of segmental damage. Investigators provided a global damage evaluation (from 0.0 to 10.0) using calculated organ damage evaluations. Predicted organ indexes and Lémann Index were constructed using a multiple linear mixed model, showing the best fit with investigator organ and global damage evaluations, respectively. An internal cross-validation was performed using bootstrap methods.

RESULTS

Data from 138 patients (24, 115, 92, and 59 with upper tract, small bowel, colon/rectum, and anus CD location, respectively) were analyzed. According to validation, the unbiased correlation coefficients between predicted indexes and investigator damage evaluations were 0.85, 0.98, 0.90, 0.82 for upper tract, small bowel, colon/rectum, anus, respectively, and 0.84 overall.

CONCLUSIONS

In a cross-sectional study, we assessed the ability of the Lémann Index to measure cumulative structural bowel damage in patients with CD. Provided further successful validation and good sensitivity to change, the index should be used to evaluate progression of CD and efficacy of treatment.