Intelliqent Accurate Quantitative Evaluation Of X-Ray Pneumothorax Based On Deep Learning

In recent years,deep learning technologies have promoted the rapid development of computer-aided diagnosis in the medical image analysis.Compared with the traditional methods,the performance of deep learning methods in image classification,detection and segmentation tasks is significantly improved,which effectively solves many problems of medical image analysis.The chest has a good contrast and is considered to be suitable part for X-ray exami-nation.Chest X-ray images are rich in radiological information,and the radiologists must carefully review every detail.In addition,with the improvement of national living standards,an increasing number of people do the regular chest X-ray examinations,which has caused a serious imbalance between the proportion of radiologists and patients.Due to the accumulation of a large number of chest radiographs,patients with critical illnesses such as pneumothorax may wait for a long time.At the same time,prolonged work overload prevents doctors from concentrating,which affects the accuracy of disease.Therefore,accurate diagnosis and quantitative analysis of pneumothorax is a difficult and urgent task for radiologists.In this paper,the intelligent quantitative evaluation of X-ray pneumothorax based on deep learning was studied.To achieve an accurate assessment of pneumothorax,it is first necessary to accurately detect pneumothorax.The pneumothorax's the shape,size,and location may change greatly,and it may overlap with other tissues.To solve this problem,a method combining dense convolutional network and gradient-weighted class activation mapping(Grad-CAM)was proposed for pneumothorax detection and localization.Because this network has the ability to integrate shallow and deep features,and the Grad-CAM algorithm is easy to visualize.Combining the two can better detect and locate the pneumothorax.And accurate segmentation is a key step in achieving pneumothorax size assessment.But the segmentation task is very challenging.Not only must we take into account the variability of pneumothorax,but also accurately segment each pixel on the image.Therefore,a full convolutional multi-scale scSE-DenseNet method for pneumothorax segmentation is proposed.This method has fewer learning parameters and low time cost.It is able to adaptively recalibrate the feature map without increasing too many parameters,while suppressing useless features and enhancing useful features.Finally,pneumothorax can be quantified using interpleural distance measurements.This article has made significant improvements to the intelligent quantitative evaluation of X-ray pneumothorax,and this technology is expected to become a pneumothorax quantitative tool for clinical application.