Impact of Noise in the Quatification of ILD Patterns in Lung CT Images
Biomedical image processing accomplishes breath taking developments and is considered integrating explorative area that has scope for research in engineering domains and medicine. Computerized problem identification of the disease has formerly evolved into decisive component in analytic and scientific trails and endeavours. Contemporary prophylactic interpretation approaches affordspromising overtures in disciplines like information and medicine as they outturn allegiance models and advances in automated form of memory and determination procedures for early detection, conclusion and treatment of cancer. The challenge lies in adequately excerpt, assess and enact the data for acquisition of morphological changes in the structural functions of body parts that are exemplified. For clinical practices, images are accumulated and stored in digital representation procured from MRI, CT, Ultrasound etc., in PACS, processing and thereby for diagnosis. Medical image segmentation is an important traipse for successive illustrations of patterns. The goal of lung image segmentation is to separate Region of Interest (ROI) for extracting lung abnormalities of Interstitial Lung Disease (ILD) impressions like Sarcoidosis, Idiopathic pulmonary fibrosis, Malignant nodules, and Sarcoidosis structures. These ROI’s are affected by background regions and exhibit various levels of quality and brightness. In this paper morphology-based segmentation is used to extract honey comb and sarcoidosis patterns that are used in as much as pronouncement and prognostication of pneumonic disease. The performance appertaining to proposed method is evaluated in noisy environment. Salt and pepper noise, Speckle noise, Poisson noise, Gaussian noise are added to original DICOM image for the evaluation of noise effect. Comparison is implemented for honey comb and sarcoidosis patterns extracted from noisy and noiseless images. Noise reduction capabilities of proposed method on a particular noise type is validated based on correlation co-efficient and peak Signal-to-Noise ratio.