Lung Cancer Detection in CT Scans of Patients Using Image Processing and Machine Learning Technique

Abstract

Detection of abnormalities in lung CT scans has always remained limited to experiences of a radiologist and capability of devices used for scanning. Again, the tremendous increase in CT data has increased the demand on radiologist’s time and brought down radiologists to patients ratio. To address this challenging problem of early detection of malignant pulmonary nodules, a set of methods were implemented sequentially over DICOM (Digital Imaging and Communication in Medicine) CT datasets of patients. These methods are broadly classified as image preprocessing and machine learning techniques. Necessary preprocessing is required for normalization and segmentation of CT datasets that are prone to external noise and composed of information about surrounding body parts. Machine learning is the process of image classification into two groups: Cancer and No-Cancer. It is implemented using convolutional neural network based on deep learning framework. CNN in a broad sense extracts features from input images and classifies the scores to either of the two classes. Analyzed results of learning are discussed in detail followed by achievements, limitations, and future scope of current work.