The results of our previous study using the hands-free manual for emergency dialysis cessation suggested a higher operational accuracy and usefulness of the hand-free manual as compared with a paper manual. However, its contents were excessively detailed and a jump function was required for speedup. Since use of a jump function that skips manual pages could cause mistakes due to false assumptions, we attempted to develop an automatic inspection method. Assuming that image recognition using a Neural Network Console (NNC) might be effective for the automatic inspection, images from a previous study in which mistakes were often made were collected and the results revealed an accuracy rate of only 37%. In order to improve the accuracy, we considered that the amount of information would need to be reduced. Therefore, we set the detection images as those showing the forceps position on the puncture needle, the relative angle between them, and the forceps grasp, and evaluated the potential merits of using a NNC, and this analysis yielded an accuracy rate of 67%. For further improvement, we conducted pre-processing, in which the level of the line connecting the centers of gravity of the markers and the midpoint of the line were moved to the center of the screen; this action yielded an accuracy of 100%, suggesting the usefulness of pre-processing.
