Intelligent judgment method for pulmonary fibrosis

Abstract

The invention discloses an intelligent lung fibrosis judgment method, which comprises the following steps of: acquiring a training sample, and pre-calibrating various types of pneumoconiosis fiber shadow pixels contained in each lung region in each CT (Computed Tomography) image chest sample image as sample labels; and inputting the training sample into the convolutional neural network model. After each CT image chest radiograph sample image is convolved in sequence, target feature vectors with different mapping scales are generated respectively. And adjusting model parameters of the convolutional neural network model to train a lung fibrosis intelligent judgment model, and carrying out lung fibrosis classification and positioning on the acquired CT image chest radiograph image. An input image in the detection module is a whole CT image chest radiograph, a feature learning model is designed to replace the last three full-connection layers with convolutional layers, and the design aims at achieving end-to-end detection of input images and output images and directly detecting focus areas of different lung areas at the same time.

Description

technical field [0001] The invention relates to the technical field of medical image analysis, in particular to a method for intelligently judging pulmonary fibrosis. Background technique [0002] The common target detection strategy is to first generate independent image blocks by moving the window slider on an entire image, use these image blocks to select target candidate areas, and then determine the target category and position, and finally integrate the target recognition area. However, there are two problems with this strategy. One is that it is necessary to use a sliding window to input the entire image into the feature extraction model, that is, the detected target features are in the image block. For the entire image, the target The region may not be complete in the image block, so there are two samples of non-target area and target area in the image block, which brings difficulty to the later image block integration. If the method is not appropriate, the integrate...

AI Technical Summary

Problems solved by technology

However, there are two problems with this strategy. One is that it is necessary to use a sliding window to input the entire image into the feature extraction model, that is, the detected target features are in the image block. For the entire image, the target The region may not be complete in the image block, so there are two samples of non-target area and target area in the image block, which brings difficulty to the later image block integration. If the method is not appropriate, the integrated target area There will be redundancy or obvious lack, that is, it cannot be accurately positioned; the second is the target detection based on the sliding window, which will bring a lot of computing overhead, because in the target image, there are only a few windows with the target area. The method is similar to the exhaustive method, which not only wastes training and testing time, but also increases the waste of storage resources of experimental equipment.

Images