Dataset Information

Comparison of Flat Colorectal Lesion Detection by Artificial Intelligence-assisted Colonoscopy Versus Endoscopists

ABSTRACT: The development of artificial intelligence (AI) systems in the field of colorectal endoscopy is currently booming, colorectal cancer being, by its frequency and severity, a real public health problem. In terms of image analysis, AI is indeed able to perform many tasks simultaneously (lesion detection, classification, and segmentation) and to combine them. Lesion detection is thus the starting point of the whole chain to choose at the end the most appropriate treatment for the patient. Large-scale studies have demonstrated the superiority of artificial intelligence-assisted detection over the usual detection by gastroenterologists, mainly for the detection of sub-centimeter polyps. However, the investigators have shown that a recent computer-aided detection system (CADe) such as the ENDO-AID software in combination with the EVIS X1 video column (Olympus, Tokyo, Japan) may present difficulties in the detection of flat lesions such as sessile serrated lesions (SSLs) and non-granular laterally spreading tumors (LST-NGs). This represents a major challenge because in addition to their shape being difficult to identify for the human eye in practice and where AI assistance would be of great value, these rare lesions are associated with advanced histology. In addition, the investigators recently described the case of a worrisome false negative of AI-assisted colonoscopy, which failed to detect a flat adenocarcinoma in the transverse colon. Therefore, it is important to measure the false negative rate of AI detection based on the macroscopic shape of the lesion. Comparing this rate to the human endoscopist’s false negatives would improve the performance of AI for this specific lesion subtype in the future.

DISEASE(S): Flat Colorectal Lesion,Colorectal Cancer

PROVIDER: 66296 | ecrin-mdr-crc |

REPOSITORIES: ECRIN MDR

ACCESS DATA

Similar Datasets

Project description:Colorectal cancer is the 2nd most common cause of death by cancer. Screening is therefore essential, with a positive impact for prevention, and in the visualization and removal of colonic adenomas, pre-cancerous lesions of colorectal cancer. The colonic adenoma detection rate (CADR), the gold standard for colonoscopy screening and screening studies, is the ratio of the number of colonoscopies with at least one histologically verified colonic adenoma to the total number of colonoscopies performed in a center. It varies between 25 and 45% depending on the center. There is a large inter-operator CADR variability, which has been linked to an increased incidence and excess mortality in colorectal cancers. To improve this detection rate, several innovative techniques have been developed: The endoscopic cap helps improve this detection rate: it is a 2cm tip with double row of plastic wings, fixed to the distal end of the colonoscope, which acts by unfolding the colonic haustrations allowing a better visualization of adenomas, and more particularly those of sessile morphology and sigmoidal location. Several multicenter studies have shown an improvement in the adenoma detection rate with this device compared to screening colonoscopy alone, with an adenoma detection rate optimization of 14%. Since then other devices, such as the Endocuff, have emerged with comparable efficacy. The Medtronic© GI GENIUS system integrates artificial intelligence (AI) to assist in the detection of polyps. It automatically identifies these precancerous lesions in real time. The study investigators previously performed a retrospective pilot study (COLODETECT), comparing colonoscopy alone as a control group, against AI alone and against the combined cap + AI. This study showed encouraging results in terms of colonic adenoma detection rate (60% for the AI + cap group versus 37% for the AI alone group versus 33% for the colonoscopy alone group, p=0.037) However, it requires a higher level of evidence to be validated in practice. This prospective COLODETECT2 study estimated an a priori expected difference between A.I. - Cap and A.I. alone of about 15% CADR. The GI GENIUS artificial intelligence system and the ENDOCUFF VISION endoscopic cap have separately proven their effectiveness in terms of colonic adenoma detection compared to colonoscopy alone. However, some limitations remain: existence of false positives (mucosal folds, residues), some morphological types still difficult to recognize (scalloped adenomas), non visualized colonic lesions. This study therefore focuses on the possible complementarity of these 2 medical devices, in order to maximize the detection rate of colonic adenomas, and thus overcome the limitations of these two techniques by optimizing the visualization of these precancerous lesions, and consequently increasing the impact of colorectal cancer screening. The study authors hypothesize that the combination of GI GENIUS AI coupled with the ENDOCUFF VISION endoscopy cap provides a better colonic adenoma detection rate (CAD) during colonoscopy than either GI GENIUS AI alone or the ENDOCUFF VISION cap alone.

Project description:Diabetic kidney disease is the leading cause of end-stage kidney disease worldwide; however, the integration of high-dimensional trans-omics data to predict this diabetic complication is rare. We develop artificial intelligence (AI)-assisted models using machine learning algorithms to identify a biomarker signature that predisposes high risk patients with diabetes mellitus (DM) to diabetic kidney disease based on clinical information, untargeted metabolomics, targeted lipidomics and genome-wide single nucleotide polymorphism (SNP) datasets. This involves 618 individuals who are split into training and testing cohorts of 557 and 61 subjects, respectively. Three models are developed. In model 1, the top 20 features selected by AI give an accuracy rate of 0.83 and an area under curve (AUC) of 0.89 when differentiating DM and non-DM individuals. In model 2, among DM patients, a biomarker signature of 10 AI-selected features give an accuracy rate of 0.70 and an AUC of 0.76 when identifying subjects at high risk of renal impairment. In model 3, among non-DM patients, a biomarker signature of 25 AI-selected features give an accuracy rate of 0.82 and an AUC of 0.76 when pinpointing subjects at high risk of chronic kidney disease. In addition, the performance of the three models is rigorously verified using an independent validation cohort. Intriguingly, analysis of the protein-protein interaction network of the genes containing the identified SNPs (RPTOR, CLPTM1L, ALDH1L1, LY6D, PCDH9, B3GNTL1, CDS1, ADCYAP and FAM53A) reveals that, at the molecular level, there seems to be interconnected factors that have an effect on the progression of renal impairment among DM patients. In conclusion, our findings reveal the potential of employing machine learning algorithms to augment traditional methods and our findings suggest what molecular mechanisms may underlie the complex interaction between DM and chronic kidney disease. Moreover, the development of our AI-assisted models will improve precision when diagnosing renal impairment in predisposed patients, both DM and non-DM. Finally, a large prospective cohort study is needed to validate the clinical utility and mechanistic implications of these biomarker signatures.

Dataset Information

Comparison of Flat Colorectal Lesion Detection by Artificial Intelligence-assisted Colonoscopy Versus Endoscopists

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure