A deep convolutional neural network for diminutive colorectal polyp recognition.
To develop a CAD-CNN system that will be able to differentiate diminutive polyps during colonoscopy with high accuracy and to compare this system to a group of endoscopist performing optical diagnosis.
Colorectal cancer (CRC) is the second most prevalent type of cancer in the Netherlands. Several studies have shown that colonoscopy is associated with a reduction in colorectal cancer mortality. This benefit is based on the detection and resection of any polyps. Diminutive colorectal polyps (1-5mm in size) have a high prevalence and very low risk of harbouring cancer (0.08%). 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, practise could become more time- and cost-effective, since histopathological examination could be omitted. In addition, patient burden could be reduced by decreasing polypectomy-associated complications and unnecessary outpatient clinic visits after colonoscopy. Studies have shown that this diagnostic optical diagnosis 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. Data comparing the diagnostic performance of CAD-CNN system to a group of endoscopists performing optical diagnosis during real-time colonoscopy are lacking.
This project is designed as a multicentre, prospective colonoscopy trial consisting of two phases. The first phase comprehends the development, training and pre-clinical testing of a CAD-CNN system. In the second phase, they will clinically validate the CAD-CNN system during real-time colonoscopies, and compare the system with the performance of endoscopists.
The outcomes of this project may lead to a more efficient and safe colonoscopies through automated CAD-CNN differentiation of colorectal polyps.
Read more about the project here.