Prediction of age and sex from paranasal sinus images using a deep learning network

摘要

Deep learning involvescomputationalmodelscomposed ofmultiple processing layers for learning representations of data withmultiplelevels of abstraction. Thisapproach has dramatically improved object detection and recognition. Thus, multi-model deep learning architecturescan contributesignificantly toward theadvancement of personalized medicine. [1] Recently,among deep learning architectures,convolutional neural network (CNN) models have demonstrated superior performancecompared to other machine-learningmethods in object detection and recognition applications. [2] Thus, these modelsarean effectivesolution forclassification and recognition problemsassociated with large datasets. In addition,compared with other learning algorithms, thelocalreceptivefieldsand shared weights ofthe CNN modelare uniquely advantageous. Therefore, the CNN modelis widely used for recognizing and differentiatingmedicalimages in clinical practice, including automated classification of gastric neoplasms based on endoscopicimages, prediction ofcardiovascular risk factors based on retinalfundus photographs,and classification ofmaxillary sinusitis based on paranasalsinus (PNS) X-ray images. [3–5] Moreover, the CNN modelfeaturesa majority decision areathat usesa classactivationmap (CAM) fromthetested dataset. Itenableseasy recognition oftheareaidentified for the performanceevaluation offeature prediction.