Developing and Implementing an Image-based Lung Abnormality Classification Approach using Contrastive Feature Extraction and RNN with Sparse Attention

Abstract

Therapeutic management of lung diseases, which ranks as the third leading cause of global mortality, holds great importance in medical diagnosis and treatment. Timely screening and diagnosis of lung abnormalities are crucial for minimizing risks by facilitating prompt and effective therapeutic interventions, which can be analyzed through digital imaging techniques. Accurately identifying the exact symptoms of lung cancer is inherently challenging because of the formation of most cancerous tissues, where large tissue structures are intersected differently. Accurate and timely detection and classification of lung abnormalities are crucial for effective diagnosis and decision-making regarding treatment planning. Recently, Deep learning methodologies have achieved superior results in medical image analysis. Therefore, this work proposes a dense-based deep learning model for classifying lung disease using medical images. Firstly, the required raw images are acquired from standard databases and used for the subsequent process. Further, the gathered images are fed as input to the Contrastive Predictive Coding (CPC) for the extraction of relevant features, which is effective at capturing high-quality latent representations from the medical images. The extracted features processed via the Recurrent Neural Network with Sparse Attention Mechanism (RNN-SA) provide classified outcomes for lung abnormalities. After the implementation, the system performance is examined and compared with other conventional methods. The accuracy of 97.47% and specificity of 98.72% for our model show that the proposed model is more efficient in classifying lung abnormalities. Therefore, the superior results are declared to prove the system efficacy in diagnosing lung abnormality.