Back to Search
Start Over
Polyp Location in Colonoscopy Based on Deep Learning
- Source :
- 2019 8th International Symposium on Next Generation Electronics (ISNE).
- Publication Year :
- 2019
- Publisher :
- IEEE, 2019.
-
Abstract
- Colorectal cancer is one of the most common cancers in China. The occurrence of most colorectal cancer is closely related to colorectal polyps. Colonoscopy is the gold standard for the diagnosis of intestinal lesions. Usually, existing colonoscopy is performed by physicians to determine the location of polyps by observing the results of detection with the naked eye. The detection rate of polyps is also affected by the doctor’s experience, fatigue, detection rate, and other factors, so there is a certain degree of polyp missed detection. Therefore, to improve diagnostic accuracy and reduce the rate of missed diagnosis, the paper proposes an improved_ssd model based on deep learning. The model is extended from the ssd_inception_v2 model, and the inception_v2 basic framework is used to extract features from multiple dimensions and fuse them, which improve the accuracy of polyp location. The test results show that the AP of this method is 94.92%, the accuracy is 96.04%, the sensitivity is 93.67%, and the specificity is 98.36%. This method realizes the accurate localization of polyps in colonoscopy and provides a reference for doctors' diagnosis.
- Subjects :
- medicine.medical_specialty
medicine.diagnostic_test
Colorectal cancer
business.industry
Deep learning
Colonoscopy
Diagnostic accuracy
02 engineering and technology
Gold standard (test)
Missed diagnosis
medicine.disease
03 medical and health sciences
0302 clinical medicine
0202 electrical engineering, electronic engineering, information engineering
medicine
030211 gastroenterology & hepatology
020201 artificial intelligence & image processing
Radiology
Artificial intelligence
Detection rate
business
Subjects
Details
- Database :
- OpenAIRE
- Journal :
- 2019 8th International Symposium on Next Generation Electronics (ISNE)
- Accession number :
- edsair.doi...........effcb125145d1cb13791aecbcb08c0c7