Dense neural network lung tumor image recognition method fusing multi-scale features

• Main Authors: TIAN JINQIN, HUO BINGQIANG, LU HUILING, DING HONGSHENG, ZHOU TAO
• Format: Patent
• Language: Chinese; English
• Published: 09.02.2021
• Subjects: CALCULATING; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses a dense neural network lung tumor image recognition method fusing multi-scale features, and the method comprises the steps: collecting and preprocessing CT modal medical images, extracting lesion ROI regions of different scales, and forming a multi-scale data set; wherein the focus ROI regions of different scales are provided with clinically marked benign or malignant tumortags; training the multi-scale data set in a dense neural network, constructing a dense neural network model, extracting a full connection layer feature vector and carrying out feature serial fusion;and obtaining a lung tumor classification result in the NSCR classifier. A dense neural network model constructed by the method is superior to an AlexNet model, bottom-layer new features can be minedagain by effectively utilizing high-layer information, transmission of the features among networks is enhanced, and feature reuse is realized and enhanced; the invention is deep in network depth, strong in network generalizati