Early detection of pulmonary vessel nodule in low dose CT image using local adaptive thresholding / Sufia Mohd Hizam, Noor Shafini Mohamad

Small lung nodules are very subtle in the medical images, and less than 30% of them are connected to the pulmonary vessels. Most of the time, the findings of this disease are not optimal at the early stages. These small lung nodules have similar HU values as pulmonary vessels, therefore, making it a challenge to separate these nodules. This study aimed to segment and suppress pulmonary vessels and detected nodules to improve the accuracy of diagnosing lung cancer by using local adaptive thresholding. This proposed framework consisted of the image enhancement process and three segmentation stages. Contrast stretching, median filter combined with closing morphological operator, and unsharp masking were employed to make the image more appealing. The first stage of image segmentation was extracting lung from the parenchyma by using a fast marching method and active contour. The second stage was to extract pulmonary vessels and nodules together using local adaptive thresholding. Extraction of the nodule from the pulmonary vessels using local adaptive thresholding was employed in the final stage. The sensitivity and specificity of this method were computed by calculating the number of pixels overlapped with the ground truth images. This proposed method presented high sensitivity and specificity for segmentation of pulmonary nodule (0.90 and 0.99) and segmentation of pulmonary vessels (0.87 and 0.99). After the suppression of the vessels, the mean CNR of the nodule increased from 3.27 to 3.61). The suppression and segmentation of pulmonary vessels in CT thorax images may reduce false-positive findings and misdiagnosis due to human error. Hence, the early discovery of lung nodules can reduce about half of the mortality rate.