DEEP LEARNING ARCHITECTURE SYSTEM FOR AUTOMATIC MEDICAL IMAGE READING

摘要

The present invention relates to a deep learning architecture system for automatic medical image reading, and more particularly, to a deep learning architecture system for automatic medical image reading, which minimizes data requirements for learning and can easily transform a deep learning architecture in a manner that is as close as possible to humans reading medical images. A deep learning architecture system for automatic medical image reading according to the present invention comprises: a trunk module (100) tying common parts together in a plurality of convolutional neural network (CNN) architectures with at least one set of feature extraction layers arranged in series consisting of a plurality of convolution layers that perform feature extraction of an image and one pooling layer that performs subsampling to reduce calculation; a branch module (200) for generating each architecture in the trunk module (100) and receiving an output of the trunk module (100) to identify a lesion in the image and diagnose a corresponding disease name; a section (110) which is an architecture in which any one branch module (200) of the plurality of branch modules (200) and the trunk module (100) are connected; and a root layer (120) for transferring an output of a specific layer of the trunk module (100) to the branch module (200) to connect the trunk module (100) and the branch module (200). The branch module (200) may be provided in plurality separately for each learned disease, and one branch module (200) and the trunk module (100) may be combined to form one section (110) for each disease, and in the case of using a new function, it is possible to perform a calculation using only the corresponding section (110) among the plurality of sections (110), thereby reducing the computational requirement and the storage requirement simultaneously.