出版社：The Institute of Image Information and Television Engineers
摘要：Cystoscopy is an examination that allows medical doctors to look inside the bladder. In examinations with a flexible cystoscope, doctors need to observe whole interior of the bladder which is broadly divided into seven parts. However, since five of the parts are similar in shape and color, it is difficult to grasp the location that the cystoscope is projecting. We have proposed the use of a system for estimating the handling patterns of a flexible cystoscope, i.e., the use of neural networks to cystoscopic images and thus give more detailed observations. However, the system could not be used to estimate the handling patterns against images which have poor characteristics. We propose a method for estimating sequential handling patterns for flexible cystoscope using prediction probabilities obtained from the neural networks. The experimental results show that the average recognition rate for estimating the handling patterns has increased by 8.3% compared with the rate of the existing method.
