Diagnostic Performance of a Convolutional Neural Network for Diminutive Colorectal Polyp Recognition

Rationale: Diminutive colorectal polyps (1-5mm in size) have a high prevalence and very low risk of harbouring cancer. Current practice is to send all these polyps for histopathological assessment by the pathologist. If an endoscopist would be able to correctly predict the histology of these diminutive polyps during colonoscopy, histopathological examination could be omitted and practise could become more time- and cost-effective. Studies have shown that prediction of histology by the endoscopist remains dependent on training and experience and varies greatly between endoscopists, even after systematic training. Computer aided diagnosis (CAD) based on convolutional neural networks (CNN) may facilitate endoscopists in diminutive polyp differentiation. Up to date, studies comparing the diagnostic performance of CAD-CNN to a group of endoscopists performing optical diagnosis during real-time colonoscopy are lacking.

Objective: To develop a CAD-CNN system that is able to differentiate diminutive polyps during colonoscopy with high accuracy and to compare the performance of this system to a group of endoscopist performing optical diagnosis, with the histopathology as the gold standard.

Study design: Multicentre, prospective, observational trial. Study population: Consecutive patients who undergo screening colonoscopy (phase 2)

Main study parameters/endpoints: The accuracy of optical diagnosis of diminutive colorectal polyps (1-5mm) by CAD-CNN system compared with the accuracy of the endoscopists. Histopathology is used as the gold standard.

The intended main result of this project is the realization of a CAD-CNN system prototype that optically diagnosis colorectal polyps during a colonoscopy with great precision and fast processing time. We hypothesize that this CAD-CNN system will be more accurate than endoscopists for making an optical diagnosis of diminutive polyps.

None

None