Sustainable framework for automated segmentation and prediction of lung cancer in CT image using CapsNet with U-net segmentation

Abstract

Due to the lung's unregulated cell proliferation, both men and women are commonly affected by lung cancer. As a result, the chest's inhale and exhale areas have substantial breathing difficulties. The World Health Organisation states that tobacco use and passive smoking are the leading causes of lung cancer. The mortality rate is still not entirely under control even though there are cutting-edge medical facilities accessible for thorough diagnosis and efficient medical care. Nowadays, machine learning has a significant impact on the healthcare industry due to its high computing capabilities for accurate data analysis and early disease prediction. So in the proposed-work, a novel classifier with an advanced segmentation approach was developed to detect lung cancer effectively. For pre-processing, image resizing, Rolling Guidance Filtering (RGF), and Switched Mode Fuzzy Median Filter (SMFMF) were used on the CT images of the lungs that were obtained from the clinic. Further, the pre-data were segmented into various groups to reduce prediction complexity using U-net segmentation. Then the segment images were utilized in a capsule neural-network (CapsNet) to detect the exact condition of the raw image. CapsNet have the capability to preserve spatial information, needing less data to train, reducing loss of features due to pooling, and quicker training period. The proposed CapsNet model provides 98% accuracy, 1.9% false positive rate and 97.9% precision. Then the performance was compared to some other existing approaches for validation. The suggested approach can detect lung cancer with low computation time, so it was well fit for real-time applications.