A novel approach to CAD system for the detection of lung nodules in CT images

https://doi.org/10.1016/j.cmpb.2016.07.031Get rights and content

Highlights

  • Grouping of nodules helped in the reduction of false positive rate and computational time.

  • Shape specific morphological operations were used for the refinement of juxtavascular nodules.

  • Intensity based, geometrical and statistical features assisted in the elimination of false positives.

Abstract

Detection of pulmonary nodule plays a significant role in the diagnosis of lung cancer in early stage that improves the chances of survival of an individual. In this paper, a computer aided nodule detection method is proposed for the segmentation and detection of challenging nodules like juxtavascular and juxtapleural nodules. Lungs are segmented from computed tomography (CT) images using intensity thresholding; brief analysis of CT image histogram is done to select a suitable threshold value for better segmentation results. Simple morphological closing is used to include juxtapleural nodules in segmented lung regions. K-means clustering is applied for the initial detection and segmentation of potential nodules; shape specific morphological opening is implemented to refine segmentation outcomes. These segmented potential nodules are then divided into six groups on the basis of their thickness and percentage connectivity with lung walls. Grouping not only helped in improving system's efficiency but also reduced computational time, otherwise consumed in calculating and analyzing unnecessary features for all nodules. Different sets of 2D and 3D features are extracted from nodules in each group to eliminate false positives. Small size nodules are differentiated from false positives (FPs) on the basis of their salient features; sensitivity of the system for small nodules is 83.33%. SVM classifier is used for the classification of large nodules, for which the sensitivity of the proposed system is 93.8% applying 10-fold cross-validation. Receiver Operating Characteristic (ROC) curve is used for the analysis of CAD system. Overall sensitivity of the system is 91.65% with 3.19 FPs per case, and accuracy is 96.22%. The system took 3.8 seconds to analyze each image.

Introduction

According to WHO report updated in 2014, the most common cause of cancer related deaths in men and women is lung cancer, accounting for 1.59 million deaths in 2012 [1]. A statistical survey by the American Cancer Society shows that death rate by lung cancer is higher than that by prostate, colon and breast cancers combined [2]. Lung cancer detection in early stage can improve the survival rate; according to Cancer Research UK, one year survival rate for patients diagnosed at stage-I is 87%, whereas survival rate for patients having lung cancer at stage-IV is 19% [3]. Because of the high sensitivity, computed tomography (CT) is generally used for the analysis of pulmonary nodule, which by definition is a small round or oval shaped growth in the lungs with diameter up to 30 mm. Presence of small pulmonary nodule can indicate that the cancer is at an early stage; different methods are used to see whether a nodule is cancerous or benign. A single chest CT scan normally produces a large number of images; each image or frame contains the information of different sections of the chest that radiologists have to analyze, which is very tedious, causing them to overlook small nodules. Computer aided detection system assists doctors in the early and efficient detection of pulmonary nodules. According to one study done by White et al., CAD system has potential to detect almost half of the lesions missed by human readers [4]. Output of CAD system acts as a second opinion for the radiologists before making final diagnosis [5].

There are three types of challenging nodules for CAD systems to detect: small nodules, ground-glass opacity nodules (GGO) and the ones attached to lung wall (juxtapleural) or blood vessels (juxtavascular). These types of nodules are often missed by CAD systems [6]. A typical CAD system analyzes chest CT scan images in three steps: lung region extraction, detection and segmentation of candidate nodules, and elimination of false positives [6].

Some researchers also add one step of preprocessing before lung region segmentation in which different image processing techniques are used for the noise removal and contrast adjustment. These techniques include Wiener filter [7], Gaussian filter [8], wavelet transform [5], [9], erosion filter [10] and histogram equalization [11]. Gomathi and Thangaraj used morphological operations for image enhancement and median filter for noise removal [12]. For contrast enhancement and noise removal, de Carvalho Filho et al. used quadratic enhancement technique and median filter respectively [8].

Some researchers divide a lung region extraction process in further steps: first segmentation of lungs takes place then lung boundaries are refined to include juxtapleural nodules. For lung region segmentation, thresholding [10], [13], [14], [15], [16] and region growing [5], [17], [18] are the most common methods used by researchers. Morphological operations [10], [17], [19], rolling ball algorithm [14] and chain code analysis [15] are generally used for contour refinement.

For the detection of pulmonary nodules, Armato et al. used multiple gray level thresholding [14]. Choi and Choi presented a CAD system which used multi-scale dot enhancement filter for nodule detection [15]. Li et al. proposed a technique of fuzzy integrated active contour model (FIACM) and parametric mixture model (PMM) for the segmentation of nodules [20].

After detection of candidate nodules, different features are extracted from them based on shape, texture and intensity, and used in the elimination of false positives with the help of classifier or on the basis of rules [5], [14]. SVM [10], [15], [20], [21] and ANN [12], [19] are the most common classifiers used in literature.

Section snippets

Materials

Data provided to the CAD system are taken from the publicly available Lung Image Database Consortium (LIDC) [22] and from Mayo Hospital Lahore, Pakistan. Data in the LIDC are provided by five academic institutions in the US. The purpose of this publicly available data is to encourage researchers to develop and assess CAD systems. Lung CT scan images of 110 patients were collected in this paper, including both men and women. A total of 11,940 slices were presented to two expert radiologists;

CAD system

Working of proposed CAD system on one case having 133 slices with slice thickness of 2.5 mm is shown step by step in Fig. 1, starting from the raw CT image input to the end result showing one true positive (TP) and two false positives (FP). The system starts with the preprocessing of raw CT scan images of chest to enhance the contrast. Next, lung region extraction from thorax is performed by intensity thresholding, and contour refinement is done using morphological closing. Combination of

Results and discussion

The main objective of the proposed system is to detect cancerous nodules from CT scan images. The system should be able to identify cancers at early stages with minimum errors. Also, analysis should be done in less time. False positive rate, sensitivity, receiver operating characteristic (ROC) curves and area under the curve (AUC) are the tools to evaluate the performance of the CAD system. Time consumption in the analysis of single slice is also essential to judge the system's performance.

Data

Research limitations

The proposed CAD system is not efficient in the detection of ground glass nodules (GGO) with low intensity values. True positive in ST group missed by the CAD system during nodule detection and segmentation stage is shown in Fig. 17A. One wall-connected tiny (WT) nodule missed during lung segmentation process can be seen in Fig. 17B. One solitary small (SS) nodule was missed in nodule detection and segmentation step.

Conclusion

A novel approach to CAD system is presented in this paper using different image processing techniques, calculating and analyzing intensity, geometric and statistical features. The system is tested on a publicly available database (LIDC), which gives the opportunity to compare the proposed system with other models tested on the same database. Results of the CAD system show good nodule detection performance as compared to other systems in literature as discussed in section 4.2. This system is

Future recommendations

Although the system's performance in the detection of most of the nodules is good as compared to other CAD systems for lung nodule detection, further effort is needed to improve the system's sensitivity for ground glass nodules having low intensity values. More features can be added in feature sets to enhance the performance of CAD system. For GGO nodules, nodule enhancement filters should be used before nodule segmentation steps. Another rewarding area of future research might be the

Acknowledgements

Authors would like to acknowledge the Lung Image Database Consortium (LIDC) and other organizations who contributed in making data available to public for research purposes. We also like to thank School of Mechanical and Manufacturing Engineering, NUST for their support in this research.

References (32)

  • C.S. White et al.

    Use of a computer-aided detection system to detect missed lung cancer at chest radiography

    Radiology

    (2009)
  • S.A. Kumar et al.

    Robust and automated lung nodule diagnosis from CT images based on fuzzy systems

    (2011)
  • M. Firmino et al.

    Computer-aided detection system for lung cancer in computed tomography scans: review and future prospects

    Biomed. Eng. Online

    (2014)
  • H. Shao et al.

    A detection approach for solitary pulmonary nodules based on CT images

    (2012)
  • X. Ye et al.

    Shape-based computer-aided detection of lung nodules in thoracic CT images

    IEEE Trans. Biomed. Eng

    (2009)
  • A. Teramoto et al.

    Fast lung nodule detection in chest CT images using cylindrical nodule-enhancement filter

    Int. J. Comput. Assist. Radiol. Surg

    (2013)
  • Cited by (124)

    • Recent advancements in deep learning based lung cancer detection: A systematic review

      2022, Engineering Applications of Artificial Intelligence
    View all citing articles on Scopus
    View full text