USAGE OF ARITIFICIAL INTELLIGENCE IN AIDING WITH COLONIC SESSILE SERRATED LESIONS DETECTION AND DIAGNOSIS (AI-SSLD)

The serrated pathway is believed to account for 30% of all colorectal cancers (CRC). However, as detection rates vary widely among endoscopists and pathologists, there is uncertainty about the prevalence of these lesions [1]. Prevalence varies with study location, diagnostic criteria and examination quality. Relatively little is known about the epidemiology of these lesions (prevalence, location, family history) and risk of malignant transformation (timing, associated factors). A systematic review reported prevalence of sessile serrated lesions (SSL) was 3.9% in Europe and 5.1% in the US.

In Asia, only few reports on Sessile Serrated lesion (SSL) have been published. In a CRC screening study from Hong Kong recruiting 6,011 subjects, 486 (8.1%) subjects were reported to have HPs and only 85 (1.4%) SSL [2]. A study from Japan recruiting 5,218 asymptomatic subjects for CRC screening reported detection rates of serrated lesions of 23.3% and of right-sided serrated lesions of 7.6% respectively [3]. In this study, high-quality video endoscopes with narrow-band imaging (NBI) and magnification were used with 0.4% indigo carmine dye to enhance the detection of flat lesions. On the other hand, In Australia, the prevalence of SSL in Chinese (2%) was lower when compared with Caucasian (7%) subjects [4]. Studies have shown that SSLs are associated with CRC, especially those on the right colon and in the elderly age group, and hence should be detected and remove.[5] In addition to the potential for malignant transformation of SSL, individuals with these lesions are reportedly at higher risk of development of synchronous and metachronous CRC and advanced colorectal neoplasia (ACN) at other sites. [6-8]

Training for endoscopists and pathologists to identify SSL will likely increase detection rates, improve the prevalence of estimates of these lesions and hence reduce the incidence of interval post-colonoscopy colorectal cancer [9] The high variability between studies on the SSL prevalence, is at least partly explained by varying detection rates of serrated lesions between endoscopists, as this rate appears highly operator dependent.

Currently, CADe has been shown to improve adenoma detection rate by around 30%.[10] With the existing algorithm, SSL detection has not been improved irrespective of endoscopist experience, system type or healthcare setting. [11] This is because focus has always been put on adenomatous polyps. SSLs are sessile or flat lesions measuring average size 5-7mm and can be easily missed during conventional colonoscopy as they are usually normal to pale in color They may exhibit distinct endoscopic features such as overlying mucus cap, cloud-like surface, ring of debris or stool around the lesion and obscured mucosal vasculature During narrow-band imaging endoscopy, they have a cloud-like appearance, irregular shape, and dark spots inside the crypts. Better bowel preparation, longer withdrawal time, and careful examination of the right colon (with repeated anterograde examination or retroflexion in the caecum) improved detection of SSL [12]. Electronic chromoendoscopy such as NBI may marginally improve the detection of SSL but is currently not recommended as mandatory practice, because clear scientific evidence is lacking.

Usage of CADe system has been shown in several studies to improve polyp detection rate even amongst junior endoscopists but however most of the CADE system is trained to focus on adenomatous polyp. This AI algorithm has been trained to detect SSL. If proven to be effective in a real-world setting, this will improve outcomes of patients undergoing colonoscopies and reduce the risk of interval colon cancers post colonoscopies.