Enhanced U-Net-based segmentation and heuristically improved deep neural network for pulmonary emphysema diagnosis.

Pulmonary emphysema is a main part of chronic obstructive pulmonary disease and lung cancer. Though, quantitative emphysema severity prediction is essential in patients with unclear lung cancer categories. Thus, diagnosis of pulmonary emphysema at the early stage is more significant and could save human life. Moreover, non-invasive positive pressure ventilation is a life-saving method that focuses on reducing the complexities in patients. When failure occurs in non-invasive positive pressure ventilation, there are more chances for mortality, which shows the significance of rational diagnosis. The delay in endotracheal intubation is avoided by developing different approaches, which leads to more demand. This paper plans to develop an enhanced system for pulmonary emphysema diagnosis using deep learning-based segmentation and classification. Initially, the detection model considers pre-processing with the help of Contrast Limited Adaptive Histogram Equalization (CLAHE) and average filtering. Moreover, lung segmentation is the major part of pulmonary emphysema diagnosis. Here, the enhanced U-Net model is utilized for lung segmentation by considering the multi-objective function with the developed algorithm. Then, the feature extraction is done using the local tri-directional weber pattern and local directional pattern descriptors. The extracted features are classified by the heuristically improved deep neural network based on a new algorithm, which will optimally diagnose the severity of pulmonary emphysema. Both segmentation and classification will be enhanced by proposing a new Electric Fish-based Grey Wolf Optimization (EF-GWO). Here, various performance metrics, like accuracy, precision, specificity, etc., on the public dataset by comparing with other models. The accuracy of the EF-GWO with E-UNet is 96%. The accuracy of the suggested developed EF-GWO based on HI-DNN is 97%. Hence, it verifies the superior performance of the recommended pulmonary emphysema disease diagnosis method with improved segmentation and classification techniques compared to other existing methods. [ABSTRACT FROM AUTHOR]