Apparatus and method for image segmentation using a deep convolutional neural network with a nested u-structure

Abstract

A non-transitory computer readable storage medium has instructions executed by a processor to receive an ultrasound image. The ultrasound image is applied to a sequence of encoders where each encoder in the sequence of encoders performs convolution neural network processing of a down-sampled version of the ultrasound image from a prior encoder, the sequence of encoders form a first dimension. The ultrasound image is applied to a transition encoder with an orthogonal dimension to the first dimension. The ultrasound image is applied to a sequence of decoders where each decoder in the sequence of decoders performs convolution neural network processing of an up-sampled version of the ultrasound image from a prior decoder, the sequence of decoders form a second parallel dimension to the first dimension. Encoder and decoder configurations and the first dimension, the orthogonal dimension and the second parallel dimension thereby define a nested U network architecture. Probability segmentation maps are produced from paired encoders and decoders in the sequence of encoders and the sequence of decoders. The probability segmentation maps are combined to form a final probability segmentation output.

Dependencies are inferred from the claim text.