Mask-RCNN detection of COVID-19 pneumonia symptoms by employing Stacked Autoencoders in deep unsupervised learning on Low-Dose High Resolution CT

This paper aim is to gathering Taiwan patients lungs chest x-ray used to AI (artificial intelligence) tech analyze lungs low-dose HRCT (High-resolution chest radiography). According to the WHO (World Health Organization) statistics pandemic records the Middle East respiratory syndrome be call to the coronavi-rus (MERS-CoV) was first detected in humans in 2012. A new novel β-coronavirus caused severe and rapidly spread to all provinces of other countries in December, 2019. The number of infected people worldwide has been exceeded 669,100, deaths 31,068, and continues to increase, end of March, 2020. The model structure with segmentation of anatomical structures on DNNs-based (deep learning convolutional neural networks) methods rely on an abundance of labeled data for proper training. AI tech can quickly to identification all X-ray images of low-dose HRCT those can detect abnormalities in patients with lab-confirmed coronavirus even before clinical symptoms appear. It’s yet key piece of evidence demonstrating the modality’s central role in containing this deadly outbreak. In addition to detecting early abnormalities, area detect HRCT revealed a large finding not seen in the latest symptom cases of COVID-19. Most noteworthy is that all COVID-19 patients in public case, according to study to the best of our knowledge, all with associated bilateral pleural effusions that is the region unlabeled HRCT dataset is more widely available be used, and necessitates approaches that yet traditional supervised learning and leverage unlabeled data for area objection segmentation tasks. The experiment deep learning algorithm model improve the existing techniques on low-dose HRCT image inspection of low-dose HRCT by deep learning structure using the segmentation function of area object detection model on Mask-RCNN. The potential of stacked autoencoder-extracted feature maps method to improve area segmentation with a DNNs structure for low-dose HRCT image. The potential of autoencoder-extracted feature maps method to improve area segmentation with a DNNs structure for HRCT. There are two parts of stacked autoencoder with DNNs based. The first part, pretrained deep convolutional stacked autoencoder with pooling-unpooling layers in CAFFE structure, and feature maps were used as initialization for the deep convolutional neural network layers in the area segmentation network. The second part,  multi-task deep learning neural network structure where the tasks of area segmentation and feature maps extraction, by means of input reconstruction, were learned net and optimized simultaneously.  The experiment result of AI tech can quickly to identification all X-ray images of low-dose HRCT those can detect abnormalities in patients with lab-confirmed coronavirus even before clinical symptoms appear. It’s yet key piece of evidence demonstrating the modality’s central role in containing this deadly outbreak. In addition to detecting early abnormalities, area detect CT revealed a finding not seen in the latest cases of COVID-19. Most noteworthy is that all COVID-19 patients in public case, according to study to the best of our knowledge, all with associated bilateral pleural effusion. This technique seeks to augment and improve existing automated low-dose HRCT for the diagnosis of bilateral pleural effusions of focus on all  COVID-19 patients with recessive. Finally, a new design AI model is proposed for a product quality improvement and shorter the clinical examination respiratory system  period check with patients .