1. Lung cancer detection and classification using deep neural network based on hybrid metaheuristic algorithm.
- Author
-
Prasad, Umesh, Chakravarty, Soumitro, and Mahto, Gyaneshwar
- Subjects
- *
ARTIFICIAL neural networks , *OPTIMIZATION algorithms , *FEATURE selection , *DATABASES , *METAHEURISTIC algorithms , *DEEP learning - Abstract
Lung cancer causes millions of deaths annually, and CT scans and lung X-rays are common diagnostic tools. However, large datasets can contain noisy and irrelevant features that can affect the effectiveness of deep learning classification systems. Feature selection is a preprocessing step that can minimize database dimensionality and improve classification accuracy by selecting essential features. In addition, to solve the disadvantage of expensive solution set analysis in the wrapper technique, this study offers an efficient evaluation methodology that significantly decreases the evaluation cost and enhances the efficacy of the feature selection method. Hence, the study proposed a Hybrid Spotted Hyena Optimization with Seagull Algorithm to solve feature selection problems, which efficiently obtained the optimum subset with the largest number of relevant attributes. Biomedical lung data are collected from the LIDC/IDRI and chest X-ray datasets, and the bicubic-interpolation approach is employed to eliminate noise present in the data. We used generative modeling technique, DCGAN, to perform data augmentation. The chosen lung characteristics are evaluated using a hybrid CNN-LSTM, which identified normal and abnormal features in biomedical lung data. The system's efficacy is determined by examining its accuracy, precision, recall, specificity, and sensitivity using a Python experimental design. In the LIDC/IDRI database, the proposed classifier achieved a sensitivity of 99.8%, specificity of 99.3%, precision of 99.14%, and accuracy of 99.6%. In the chest X-ray dataset, the classifier achieved a sensitivity of 99.62%, specificity of 97.8%, precision of 97.5%, and accuracy of 99.7%. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF