Validation of an Artificial Intelligence System Based on Raman Spectroscopy for Diagnosis of Gastric Premalignant Lesions and Early Gastric Cancer

Overview

Early detection and treatment of gastric premalignant lesion and early gastric cancer (EGC) have been proposed to improve outcomes of gastric cancer. Gastric dysplasia is a premalignant lesion and the penultimate stage in gastric carcinogenesis. On white light endoscopy (WLE), it is difficult to distinguish gastric dysplasia and EGC from benign pathology such as gastric intestinal metaplasia (GIM). Image enhanced endoscopy such as narrow-band imaging (NBI) is recommended to improve characterization of suspicious gastric lesions detected on WLE. Magnified-endoscopy with NBI (ME-NBI) have been shown to be superior to HD-WLE for diagnosis of GIM and EGC. Data on gastric dysplasia is less robust. Ultimately, biopsy is required to confirm diagnosis of gastric dysplasia/EGC. Gastric dysplasia can be classified into low-grade dysplasia (LGD) or high-grade dysplasia (HGD). Biopsy sampling may not be representative of the final histopathological grade of resected specimens and may under-stage dysplasia. Thus, endoscopic resection (ER) is recommended for gastric dysplasia and EGC on biopsy for diagnostic and therapeutic purpose. The current gap is to improve concordance of endoscopic and histologic findings of gastric dysplasia and early gastric cancer. Raman spectroscopy based artificial intelligence system (SPECTRA IMDx) was developed to provide an objective method to identify patients with gastric premalignant lesions and EGC. SPECTRA IMDx interrogate tissues at the cellular level and utilizes molecular information to provide actionable information to endoscopist during gastroscopy. Studies on diagnostic performance using Raman spectroscopy analysis devices have shown high sensitivity and specificity in detection of gastric cancer and precancerous lesions compared to WLE. However, these studies included few GIM, gastric dysplasia and gastric carcinoma. It is still unclear if Raman spectroscopy outperforms WLE in diagnosis of gastric HGD and EGC. In addition, the Raman spectroscopy algorithm is only able to characterize lesions into high risk (HGD/EGC) versus low risk (GIM/LGD/Gastritis/Normal). It is also uncertain if this technology is able to differentiate GIM and LGD. We plan to conduct a prospective trial to validate the diagnostic accuracy of SPECTRA for prediction of gastric HGD and EGC prior to gastric ER. Hypothesis: SPECTRA IMDx is able to differentiate higher risk lesions (HGD/EGC) from lower risk tissue/lesion (GIM/LGD/Gastritis/Normal)

This is a prospective, single-centre study designed to evaluate the diagnostic performance of SPECTRA IMDx for real-time characterisation of gastric dysplasia and early gastric cancer for patients undergoing gastric endoscopic resection over a 24-month period. Informed consent will be taken by the investigator before study enrolment. Subject participation in this study will last only the duration of the gastric endoscopic resection procedure and during surveillance gastroscopy. The duration of procedure may vary depending on the technical difficulty of the endoscopic resection performed, the size and number of suspicious lesions detected requiring biopsies. Subjects will not be required to make an additional visit to the hospital for the study. Pre-endoscopic preparation and endoscopic procedure All antiplatelets except aspirin and all anticoagulants will be stopped before gastric endoscopic resection as per guideline.Full blood count and coagulation panel will be checked within one month before procedure to ensure no significant thrombocytopenia or coagulopathy. All patients on antiplatelets except aspirin and anticoagulation will be instructed to stop medications before procedure to reduce risk of bleeding. All patients will be fasted and receive oral mucolytics prior to gastroscopy as per routine practice. Intravenous antispasmodics with either buscopan or glucagon will be administered to allow proper assessment of lesion prior to resection. The Endoscopist will begin systemic examination of the stomach to inspect for presence of concomitant suspicious lesions and to identify the target lesion with dysplasia for endoscopic resection. Once target lesion for ER is identified, the lesion is examined with HD-WLE followed by NBI. The endoscopist will then classify the lesion as normal/Gastritis/GIM/LGD or HGD/early gastric cancer based on HD-WLE and NBI. The endoscopic diagnosis is documented on the case report form. Following that, SPECTRA IMDx probe will be inserted via the instrument channel and placed in contact with the lesion first for evaluation of lesion followed by collection of Raman spectroscopy signals. The SPECTRA IMDx system will detect the scattering light signal from the lesion and classify the lesion as 'high-risk' or 'low-risk'. If there are different risk classified on the same lesion, the highest grade of pathology will be recorded. This result will be documented on case report form. This will be followed by collection of Raman signals from gastric lesion. Workflow for SPECTRA IMDx usage is divided into 1. Evaluation of lesion for comparison against histopathology: 2. Collection of Raman spectroscopy signals to differentiate between GIM, LGD, HGD and EGC If additional suspicious lesions are identified during systematic examination of the stomach which were not biopsied or detected during previous endoscopy, these lesions will be first examined with HD-WLE followed by NBI. SPECTRA IMDx will be then be used to classify the lesions before biopsy sampling/resection are carried out. Findings from HD-WLE with M-NBI and SPECTRA IMDx will be documented in CRF as per above protocol. For lesions that does not require resection, collection of Raman spectroscopy signals will be collected by placing SPECTRA IMDx probe on lesion for 5 seconds at point of biopsy. Following assessment of gastric lesions with SPECTRA IMDx, indigo-carmine chromoendoscopy will be applied onto the lesion to delineate the lesions prior to endoscopic resection. Endoscopist will document findings based on indigo-carmine chromoendoscopy on the case report form. Endoscopic resection is then performed. If the lesion is deemed not suitable for endoscopic resection due to size criteria or no evidence of dysplasia based on HD-WLE and NBI with dye-based chromoendosocpy, biopsies will be taken instead as per routine practice. Any adverse events related to gastric endoscopic resection will be noted. The time taken for examination using HD-WLE and NBI, gastric endoscopic resection and overall procedure time will be recorded. Surveillance endoscopy is usually performed within six months to a year following gastric endoscopic resection to ensure no recurrence. During surveillance endoscopy, endoscopic resection site and any suspicious lesions will be examined sequentially with HD-WLE, M-NBI and lastly SPECTRA IMDx (first for evaluation of lesion, followed by collection of Raman spectroscopy signals. Findings will be documented in CRF before biopsy sample(s) are taken. For all patients, resection specimens (or biopsy specimens in cases where endoscopic resections are not performed) will be fixed in formalin and sent for histopathological evaluation. A specialized gastrointestinal pathologist will examine the resection specimen and/or biopsy sample(s) and provide a histopathological diagnosis for the lesion as part of patient routine care. The result will be recorded in case report form. The reporting pathologist will be blinded to SPECTRA IMDX and HD-WLE with M-NBI assessment. The endoscopic diagnosis of SPECTRA results and HD-WLE (with and without NBI) will be compared against the final histopathological results. Acquired Raman spectroscopy signals will be also be analysed to improve diagnostic algorithm of SPECTRA in differentiating GIM, LGD, HGD, ECG and scar tissue after correlation with histopathological specimen.