Diagnosing lung disorder and optimal classification using dense convolution (Dc)-CapsNet technique

Lung carcinoma is regarded as lung cancer and is the usual form of cancers worldwide. It cannot be diagnosed easily until it spreads and is regarded as a complex concern to treat. To overcome existing issues, there is a need to detect and classify such lung disorder at an early stage by predicting normal and malignant nodes. This aids in providing risk factor of lung carcinoma for the patients and saving lives with necessary treatment. For this purpose, a new scheme is developed for diagnosing lung disorder and optimal classification of lung disorder with the use of Dense Convolution (DC)-CapsNet model. This framework's primary goal is to use chest X-ray and CT scan pictures to identify and categorize a variety of lung conditions, including lung cancer, pneumonia, and tuberculosis. Here, we suggested a brand-new Deep Learning method for lung illness identification. Image preprocessing, segmentation, feature extraction, feature optimization, and detection are the steps that are involved. This study presents a deep learning-based automated image-based approach for lung disease diagnosis and classification. This work's primary contribution is as follows: Optimal filtering without data loss is used in the pre-processing step to improve the quality of the chest X-ray image. Augmented Profuse Clustering (APC) is used to perform the segmentation. The Deep CNN model serves as the basis for the feature extraction procedure. Then, in order to achieve the best feature, the best features are chosen using the Novel Evolutionary Water wave optimization algorithm. After that, the DC-CapsNet model is used to predict and detect lung diseases. This method determines whether or not the input lung image is normal. Finally, the assessment of performance is carried on proposed model and the outcomes obtained will be compared with various traditional models to validate the enhancement of proposed scheme over other compared schemes. Numerous performance parameters, including overall accuracy, precision, recall, F1-score, specificity, sensitivity, processing time vs. training time, Jaccard similarity coefficient (JSC) and Dice similarity coefficient (DSC), and validated efficiency, were used to assess the performance. The accuracy is improved to 97.58%, outperforming existing models compared by 1.9% and attains high precision rate of 0.986, recall as 0.975, F1-score of 0.972, DSC of 0.995, and JSC (0.954) among entire models compared. The assessment reveals that presented model offers enhanced outcome than existing models in all metrics estimated thus revealing the efficiency of proposed model in predicting lung disorder.