semantic segmentation under the condition of the ultrasonic ima
ge segmentation task for lung nodules. Implementation is done
through python technology and deep learning of the Tensorflow
framework.
Based on semantic segmentation technology of deep learning, thi
s paper modifies and designs a network model based on semanti
c segmentation of deep learning, which can learn convolutional
neural networks by referring to network architecture of backg
round matting. In the network model input of a medical visual
image that is difficult for human to distinguish, The locati
on of interest can be quickly and efficiently distinguished, a
nd the model outputs its alpha map, thus making efficient med
ical diagnosis possible.
On the surface of the experiment, the AUC of this model reac
hed 0.96, which exceeded the effect of FCN semantic segmentati
on network in 2015, and the speed was also higher than that
of general semantic segmentation technology. The model was tr
ained in GPU mainly through marked images, and the network in
put three-channel medical image. The network outputs the alpha
diagram of channel 1 to continually learn and update the pa
rameters of the iterative neural network. Finally, the network
converges at the global optimum, and the loss function is r
