Abstract

Cancer is hard to analyze in light of its convolution. It is a heterogeneous disease which adds to the trouble of conclusion and forecast. The lung tumor is among the most perpetrating kind of malignancy. It has high occurrence rate and high death rate as it is frequently analyzed at the later stages when it is trying to treat it. Like this, a noteworthy research exertion is being made to help the oncologists in early lung disease diagnosis. Computed tomography (CT) scans are broadly used to distinguish the disease; it envisions little nodules or tumors which cannot be seen with a plain film X-ray. Personal computer helped gadgets are being created to analyze the ailment at prior stages productively. In this paper, we present a fully automatic framework for nodule detection from lungs CT scans. The working of the proposed system can be divided into two phases: lung segmentation and nodule detection. The preparatory phase of lung malignancy analysis using computer-aided design is lung segmentation from the chest CT scans. Accurate lung segmentation is momentous in such frameworks as the execution of the later stages in such examination to a great extent relies upon the division accuracy. In the proposed system, lung segmentation starts with the pre-processing of the CT scans which incorporates the transformation of DICOM pictures into a loss aversion PNG format. A histogram of the grayscale CT image is built to automatically evaluate the limit to isolate the lung locale from the foundation. Then the associated segments are processed to expel any residual foundation. The morphological operators are utilized to enhance the segmentation accuracy. In the second phase, the internal structures are extracted from the parenchyma. A threshold-based technique is proposed to separate the nodules candidates from the other structures such as bronchioles and blood vessels. Different statistical features and shape-based features are extracted for these nodules candidates to formulate a nodule feature vector. These features are then classified using support vector machines. The proposed method is evaluated on a large LIDC dataset achieving a sensitivity rate of 93.75% and 0.13% false positive rate per image which shows its effectiveness.