Improved Heuristic Algorithm-Assisted Residual DenseNet with Transformer-Based Unet Segmentation Model for Pulmonary Emphysema Disease Detection
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Permanent deterioration of the lung’s parenchyma is a characteristic of pulmonary emphysema, also known as Chronic Obstructive Pulmonary Disease (COPD), and is a disability and primary cause of death in developed nations. Emphysema represents a key contributing factor to this condition. The early identification of this condition is therefore more important and may even save lives. Furthermore, a life-saving technique is significant that aims to simplify patient care is non-invasive controlled breathing. Hence, this work aims to employ the deep learning model to diagnose the disorder effectively and significantly. First, the required Computed Tomography (CT) images are collected from publicly available datasets: local clinical cohort and Computed Tomography Emphysema Database. Then, the cleaning task is done using the Contrast Limited Adaptive Histogram Equalisation (CLAHE) and bilateral filtering, where the pre-processed image is achieved. Subsequently, the resultant images are fed into a segmentation network named Transformer Recurrent Residual U-net (TR2Unet). Finally, the segmented region is subjected to the classification model of Adaptive Residual DenseNet (ARes-DNet), in which the hyper-parameter optimisation is done by using an Improved Mother Optimisation Algorithm (IMOA). Hence, the newly recommended model is examined with various measures such as dice coefficient, accuracy, IoU, jaccard coefficient, precision and NPV, demonstrating superior performance in both segmentation and classification operations. Thus, this framework provides a promising solution for faster, highly accurate emphysema diagnosis, greatly improving patient outcomes via earlier detection and focused treatment. The proposed model attains 98.5% accuracy while analysing with dataset-1 and similarly, higher performance has been achieved in dataset-2 analysis with 98.7%.
Fields of Study & Subjects
Medical Image Segmentation Techniques
COVID-19 diagnosis using AI
Radiomics and Machine Learning in Medical Imaging
Publication details
- Journal / source
- Journal of Information & Knowledge Management
- Volume
- 25
- Issue
- 08
- Pages
- 2650015
- Publication date
- 31 March 2026
- Language
- EN