Using Convolutions and Image Processing Techniques to Segment Lungs from CT Data

In this paper, we put forward a novel approach of lung segmentation using convolutional neural networks (CNNs). CNNs have outperformed traditional techniques in many visual recognition tasks over the years. In the traditional methods, the features of the images to be recognized and then segmented are often hand-coded. This might seem quite simple at first and easy to implement but then as the complexity of the problems increases, these methods start to crumble. Moreover, these methods do not handle edge cases accurately as the programs do not know what to do when the images deviate from the expected nature. This is where deep learning methods (CNNs in our case) come to the rescue. Instead of hard-coding, we have built a CNN which is fed with a dataset consisting of CT images and the segmented images of the lungs. CNN automatically detects the features, and the accuracy is increased by tuning the parameters as well as using other techniques. The model once trained is then tested using the test dataset. As evident from the results, this approach gives a much higher accuracy compared to traditional methods, ultimately paving the way to a better analysis of the lungs.