In vivo evaluation of complex polyps with endoscopic optical coherence tomography and deep learning during routine colonoscopy: a feasibility study

Abstract

Standard-of-care (SoC) imaging for assessing colorectal polyps during colonoscopy, based on white-light colonoscopy (WLC) and narrow-band imaging (NBI), does not have sufficient accuracy to assess the invasion depth of complex polyps non-invasively during colonoscopy. We aimed to evaluate the feasibility of a custom endoscopic optical coherence tomography (OCT) probe for assessing colorectal polyps during routine colonoscopy. Patients referred for endoscopic treatment of large colorectal polyps were enrolled in this pilot clinical study, which used a side-viewing OCT catheter developed for use with an adult colonoscope. OCT images of polyps were captured during colonoscopy immediately before SoC treatment. A deep learning model was trained to differentiate benign from deeply invasive lesions for real-time diagnosis. 35 polyps from 32 patients were included. OCT imaging added on average 3:40 min (range 1:54-8:20) to the total procedure time. No complications due to OCT were observed. OCT revealed distinct subsurface tissue structures that correlated with histological findings, including tubular adenoma (n = 20), tubulovillous adenoma (n = 10), sessile serrated polyps (n = 3), and invasive cancer (n = 2). The deep learning model achieved an area under the receiver operating characteristic curve (AUROC) of 0.984 (95%CI 0.972-0.996) and Cohen's kappa of 0.845 (95%CI 0.774-0.915) when compared to gold standard histopathology. OCT is feasible and safe for polyp assessment during routine colonoscopy. When combined with deep learning, OCT offers clinicians increase confidence in identifying deeply invasive cancers, potentially improving clinical decision-making. Compared to previous studies, ours offers a nuanced comparison between not just benign and malignant lesions, but across multiple histological subtypes of polyps.

Keywords: Colonoscopy; Colorectal cancer; Deep learning; In vivo; Optical coherence tomography; Polyp.

Fig. 1

Representative images of non-invasive pre-malignant lesions (TA, TVA, SSP) and malignant (cancer) lesions under white light endoscopy, OCT (radial and rectangular zoomed), and hematoxylin & eosin (H&E) staining from fourin vivo lesions. In OCT images, the layer above tissue is the PTFE tubing. The radial OCT image includes a green bounding box showing the approximate location of the rectangular zoomed OCT image (a) TA demonstrates elongated vertical teeth patterns, indicating the presence of tubular structures observed in the H&E images. (b) In TVA, the presence of villous structures shows on OCT as elongated finger-like projections projecting in all directions. (c) SSP generally displays a thinner mucosa layer with patterns of high intensity without the characteristic teeth patterns, and the submucosa layer is often visible. (d) Malignant lesions lacked distinct teeth or layered structures, presenting irregular shapes with low contrast and are relatively homogenous compared to other lesions. Note: in white light images, the ring around the image is the distal attachment cap commonly used in colonoscopy to facilitate observation and maintain a clear field of view.

Fig. 2

Benign vs. malignant binary classification results. (a) Predicted probability of cancer on the test set, stratified by histology type. (b) ROC curve for cancer vs. benign on the test set with AUC score of 0.984 and 95% confidence interval of 0.972–0.996.

Fig. 3

Portable endoscopic OCT imaging system. (a) Focusing assembly at the tip of the fiber optic OCT probe. The surface of the GRIN lens is polished at ~ 9 degrees. (b) Close-up photograph of the encapsulated probe tip. (c) Assembled OCT system. The laser, optical components, and PC are housed in different levels within the cart. The rotary joint and the probe are mounted on a monitor arm extending from the rear of the cart.

Fig. 4

OCT system resolution and depth. (a) Lateral resolution of the OCT catheter in the x direction. (b) Lateral resolution in y direction. (c) B-scan image of a segment of Scotch™ tape wrapped around OCT probe. 28 tape layers can be discerned, corresponding to a 1.4-mm imaging depth in tape.

Fig. 5

Deep learning image classification strategy. (a) ViT model architecture for transfer learning. Images are projected into patches, combined with a positional embedding, and passed through the transformer encoder for feature extraction. A special CLS token is appended to the patch embeddings as the classification token. The transformer encoder output from the CLS token is passed through the MLP classification head. The output from the MLP is converted to the predicted probability of a class with a softmax activation function. (b) Transfer learning pipeline. The ViT model pretrained on ImageNet using self-supervised learning is fine-tuned on ex vivo cancer vs. in vivo benign complex polyp data and evaluated on in vivo only data.