Auto Diagnostics of Lung Nodules Using Minimal Characteristics Extraction Technique

Abstract

Computer-aided detection (CAD) systems provide useful tools and an advantageous process to physicians aiming to detect lung nodules. This paper develops a method composed of four processes for lung nodule detection. The first step employs image acquisition and pre-processing techniques to isolate the lungs from the rest of the body. The second stage involves the segmentation process using a 2D algorithm to affect every layer of a scan eliminating non-informative structures inside the lungs, and a 3D blob algorithm associated with a connectivity algorithm to select possible nodule shape candidates. The combinations of these algorithms efficiently eliminate the high rates of false positives. The third process extracts eight minimal representative characteristics of the possible candidates. The final step utilizes a support vector machine for classifying the possible candidates into nodules and non-nodules depending on their features. As the objective is to find nodules bigger than 4mm, the proposed approach demonstrated quite encouraging results. Among 65 computer tomography (CT) scans, 94.23% of sensitivity and 84.75% in specificity were obtained. The accuracy of these two results was 89.19% taking into consideration that 45 scans were used for testing and 20 for training. The rate of false positives was 0.2 per scan.

3D pulmonary nodules detection using fast marching segmentation

“…This is also similar to the work proposed by [18] that segmentation is carried out in 2D environment whilst feature extraction is performed in 3D environment. In difference and extras, this study has proposed three main contributions; (i) a new segmentation method for lung parenchyma extraction, (ii) 3D…”

“…In this method, seed locations are defined by a global threshold value while pixel weights are calculated based on the difference between each pixel's intensity and the average intensity of all pixels. Although the fast marching technique introduces these additional steps, it has eased the selecting of a proper threshold value that generally required in [18]. In additions, edge detection and morphological operations were borrowed to solve the border reconstruction problem which is always a case in juxta-pleural nodule segmentation.…”

“…In this work, we considered the random forest classifier because it can improve the classification accuracy by using ensemble multiple decision trees. Although SVM in [18] can give the high accuracy of classification, it depends on the optimal choice of kernel. …”

Lung Nodule Detection in CT Images Using Statistical and Shape-Based Features

“…2 . Lung in a CT scan 12 (i) Background (ii) Dark gray circular region (iv) Lungs in dark grey shade (iii) Brighter region To reduce the image to our ROI, we perform thresholding. Thresholding is mainly dependent on the value of the threshold, and this value is usually user specified.…”

“…False Positives (FN) Total number of tested images (12) Similarly, the false positive per exam (FPE) is the ratio of false positives to the total number of cases evaluated in the experiment. The statistics presented in Table 2 reveal that our method achieved convincing results.…”