Deep Neural Network-Based Ct Scan Analysis For Lung Disease Detection In Diabetic Patients

Abstract

Lung disease diagnosis in diabetic patients poses a significant challenge due to the complex and overlapping patterns observed in CT scan images. These scans often exhibit a wide range of pulmonary surface abnormalities, making it difficult for radiologists to distinguish between various lung diseases accurately. This research proposes a novel deep learning-based approach for the automated detection and classification of lung diseases in diabetic individuals using Deep Neural Networks (DNN). The study focused on the identification of diabetic stages using fundus images and integrates this diagnostic insight with pulmonary analysis to improve disease detection efficiency.The proposed system employs a DNN-based classifier trained on annotated CT scan images that have been validated by certified radiologists. The classifier not only analyzes lung images for disease patterns but also correlates them with the severity of diabetes, offering a comprehensive diagnostic model. Uniquely, the system takes the fundus image of a diabetic patient as an input to estimate the diabetes level and, based on that, predicts the likelihood and extent of associated lung diseases. This approach enhances early detection and allows for disease stratification based on diabetic stages. The model demonstrates high accuracy in detecting a variety of lung diseases commonly seen in diabetic patients, including asthma, pneumothorax or atelectasis, bronchitis, chronic obstructive pulmonary disease (COPD), lung cancer, and pneumonia. Its rapid image processing and low classification error make it suitable for large-scale, real-time screening applications. Designed with mass screening in mind, this DNN-based classifier has the potential to assist clinicians and healthcare providers in early intervention and treatment planning, particularly in resource-constrained settings where specialist radiologists may not be readily available.