Deep learning architecture system for automatic fundus image reading and automatic fundus image reading method using deep learning architecture system

A deep learning architecture system for automatic fundus image reading, the system comprising:

• a trunk module that combines common parts in a plurality of convolutional neural network (CNN) architectures into one part, the CNN architecture having one or more serially arranged feature extraction layer sets respectively configured of a plurality of convolutional layers for performing feature extraction on a fundus image and one pooling layer for performing subsampling to reduce the amount of computation;
• a plurality of branch modules for receiving an output of the trunk module that has generated an architecture for each of the branch modules, identifying a lesion from the fundus image, and diagnosing a corresponding disease name;
• a section which is an architecture that connects any one branch module among the plurality of branch modules to the trunk module;
• a root layer for connecting the trunk module and the branch module by transferring an output of a specific layer of the trunk module to the branch module;
• and
• a final diagnosis unit for integrating diagnosis data received from the plurality of branch modules to determine and output a final disease name, wherein
• the branch module includes:
• a disease inference unit for inferring a disease category by generating a learning data using a matched disease name or disease code in a doctor's charting process;
• a location search unit for finding an optic nerve head (ONH) from the fundus image, classifying a blind spot ratio (vertical cup-to-disc ratio, VCDR) corresponding thereto, and searching for locations of an optic nerve disc and a macula;
• a key lesion finder for finding a key lesion that is a component constituting a disease;
• and
• a small-sized lesion finder for detecting a small-sized lesion from the fundus image.