Identification of a small, depressed type of colorectal invasive cancer by an artificial intelligence-assisted detection system

A 64-year-old man underwent surveillance colonoscopy with a computer-aided detection (CADe) system (EndoBRAIN-EYE; Cybernet Systems, Tokyo, Japan) [1]. The system identified a 5-mm slightly reddish lesion in the sigmoid colon. Spraying with indigo carmine enabled identification of a clearly depressed area on the lesion ([Fig. 1], [Video 1]). The lesion showed type V_I pit pattern, indicating high grade dysplasia or slightly invasive submucosal cancer [2]. Endoscopic mucosal resection was performed. Pathological examination showed a well-differentiated adenocarcinoma with slight invasion of the submucosal layer ([Fig. 2], [Fig. 3], [Fig. 4]).

Fig. 1 Spraying of indigo carmine revealed a clear depression (Paris 0-IIc lesion).

Video 1 The EndoBRAIN-EYE (Cybernet Systems, Tokyo, Japan) outputs bounding boxes of suspected polyp candidate areas. The left image is the system’s output, and the right image is the original endoscopic image.

Fig. 2 Crystal-violet dye staining with magnification, showing irregular-shaped pit patterns of varying sizes (Kudo’s type V_I pit pattern).

Fig. 3 Photomicrograph of the specimen (hematoxylin and eosin staining), showing a well-differentiated adenocarcinoma.

Fig. 4 Photomicrograph showing that one cancerous gland (red arrow) invaded the submucosal layer beyond the muscularis mucosa (desmin immunostaining).

Artificial intelligence (AI) technology has regulatory clearance and is increasingly used during colonoscopy. A meta-analysis showed that CADe systems increase adenoma detection rates [3]. However, identifying subtle nonpolypoid lesions (e. g. 0–IIc type depressed lesions; laterally spreading tumors without granules) with CADe is still considered challenging. This is clinically relevant because a recent randomized trial found that such nonpolypoid tumors may be one of the causes of post-colonoscopy colorectal cancer [4]. Such lesions have greater malignant potential than other tumor morphologies and are often overlooked because of their appearance [5]. To the best of our knowledge, this is the first report of detection of a depressed, type 0–IIc lesion by CADe in real time during clinical colonoscopy. This AI-assisted detection was of particular value because the lesion was found to be a submucosally invasive colorectal cancer.

Endoscopy_UCTN_Code_TTT_1AQ_2AB

Publication History

Article published online:
21 December 2021

• References

• 1 Ishiyama M, Kudo SE, Misawa M. et al. Impact of the clinical use of artificial intelligence-assisted neoplasia detection for colonoscopy: a large-scale prospective, propensity score-matched study (with video). Gastrointest Endosc 2021; DOI: 10.1016/j.gie.2021.07.022.
• 2 Kudo SE, Rubio CA, Teixeira CR. et al. Pit pattern in colorectal neoplasia: endoscopic magnifying view. Endoscopy 2001; 33: 367-373
• 3 Barua I, Vinsard DG, Jodal HC. et al. Artificial intelligence for polyp detection during colonoscopy: a systematic review and meta-analysis. Endoscopy 2021; 53: 277-284
• 4 Matsuda T, Fujii T, Sano Y. et al. Randomised comparison of postpolypectomy surveillance intervals following a two-round baseline colonoscopy: the Japan Polyp Study Workgroup. Gut 2020; 70: 1469-1478
• 5 Soetikno RM, Kaltenbach T, Rouse RV. et al. Prevalence of nonpolypoid (flat and depressed) colorectal neoplasms in asymptomatic and symptomatic adults. JAMA 2008; 299: 1027-1035