| Popis: |
The segmentation procedure of thoracic CT images is essential to produce a systematic diagnosis and disease detection. There are many healthcare facilities and hospitals available worldwide where manual analysis of medical images is being performed. An increasing amount of medical images, such as thoracic CT images, making this manual procedure much complex, inefficient, and prone to error. This situation can be easily overcome by utilizing a computer-automated detection (CAD) system, which is suitable to generate an image analysis with a reduced error percentage. However, it requires an efficient segmentation technique to produce better reliable outcomes. The notion of this article is to delineate the efficient lung segmentation technique. This article describes the Convolutional neural network (CNN) based Lung nodules detection methodology. CNN is used here to learn the knowledge from a large amount of data, in order to provide improved detection mechanisms. An advanced SumNet based architecture has been incorporated along with CNN network for reducing training complexity by enhancing the learning rate. The performance score of the proposed method shows 94.11% accuracy which is more relevant for the image analysis as compared to other existing techniques. It has also achieved 0.94 score for specificity and 0.93 dice score. The training losses are also reduced to 0.10 for training and 0.05 for validation operation. After introducing SumNet architecture, it has been found that the speed of operations is improved in terms of training, testing, and validation performance. It further allows getting insights from the vast amount of data in less amount of time. This process delivers improved output parameters for lung nodule detection, and certainly, it would be preferred for the development of better CAD systems. |
