Fractal Analysis Of Lung CT Images Based On Fractal Characteristic Parameters

The COVID-19 outbreak has had a dramatic impact on every aspect of daily life,and the disease has become one of the deadliest pandemics in human history.At present,COVID-19 testing is mainly done through nucleic acid testing.Although this method takes a short time,it requires multiple tests to ensure the accuracy of nucleic acid results,which poses a great challenge to the epidemic prevention work.This paper proposes a method to detect COVID-19 based on lung CT images.Medical CT image has been widely used in the medical field because of its advantages such as high sensitivity of detection results,simple operation steps and non-trauma to the human body.Medical staff can clearly observe the pathological sites of organs with the help of CT image.This paper attempts to apply fractal theory to the analysis of lung CT images,so as to accurately identify patients with COVID-19.In the process of image preprocessing,lung parenchyma segmentation and Haar wavelet transform were performed.Since the lesion location of COVID-19 mainly occurs in the lung parenchyma,separate extraction of the lung parenchyma can greatly improve the classification accuracy,and then describe the texture features of lung CT images by fractal feature parameters.In order to improve the accuracy of classification,wavelet transform and gray co-occurrence matrix are introduced in this paper.The gray co-occurrence matrix method can quantitatively analyze the image texture,and wavelet transform can segment an image into different sub-images according to different frequencies,which is helpful to extract more information from the image.Random forest algorithm and Bayesian optimization are used in feature selection and model training respectively.Random forest feature selection is mainly to reduce the redundancy among features,eliminate the insignificant feature variables and eliminate the correlation between features.Bayesian optimization was used to find the optimal parameters of each model.After comparison,it was found that the ensemble learning model had the best effect.After adding fractal features,the classification accuracy of random forest model was 93.2%,and that of Lightgbm model was 94.1%.Finally,a Stacking model was applied to integrate the well-performing single models,achieving classification accuracy of 95.0%.This model can well recognize lung CT images,which can be used to quickly determine whether a patient has COVID-19,which is of great significance in disease detection.In practical application,it can be used as an auxiliary means for pharyngeal swab detection to improve the accuracy of results and reduce risks.In addition,the model can also be used as an objective auxiliary diagnostic method for the detection of other lung diseases and other tissue diseases to reduce the probability of misdiagnosis.