Virtual Library

Start Your Search

Y. Xie

Author of

  • +

    MS 24 - CT Screening: Minimize Harm/Cost and Risk Assessment (ID 42)

    • Event: WCLC 2015
    • Type: Mini Symposium
    • Track: Screening and Early Detection
    • Presentations: 1
    • +

      MS24.02 - Computer Assisted Lung Cancer Screening: Automated CT Image Analysis (ID 1955)

      14:15 - 15:45  |  Author(s): Y. Xie

      • Abstract
      • Presentation

      With the advent of lung cancer screening (LCS) with low-dose chest CT images, the attention for computer aided tools advances from proof of concept and validation studies to clinical utility. Computer aided image-based diagnosis tools (CAD) for LCS on the initial CT have a primary objective of improved decision making for follow up actions. There are four roles for CAD tools in this context: nodule detection, nodule characterization, nodule growth-rate measurement for malignancy status, and companion diagnostics. The special low-dose CT scan acquired as the primary test in LCS is of lower quality than a traditional clinical CT scan and, consequently, presents a higher challenge to computer analysis methods. Computer aided nodule detection systems address the critical screening task of identifying pulmonary nodules in low-dose CT images. These systems typically identify the location of nodule candidates in the CT images. In general, they detect small sphere like high intensity image regions that correspond to the most common and important finding in LCS. Their performance is related to size and most evaluations are focused on nodules of 4-5 mm or larger. For smaller nodules the false positive rate is much higher. The first of such systems received FDA approval in 2004. There has been significant technology improvement since then with sensitivities in research systems higher than 90% reported in 2007 [1]. In 2012 Zhao et al [2] reported on a study using commercial software on 400 randomly selected cases from the NELSON study. They found that the CAD system could obtain 96.7% sensitivity on nodules greater than 50 mm[3] (4.6 mm) with only 1.9 false positives per scan. In contrast, the double reading achieved 78.1% sensitivity. While the benefit of using computer detection for LCS has been clearly demonstrated and good commercial products are available, there has been little adoption of these methods in recent LCS studies. The second area in which the computer may by useful is in analyzing the images of pulmonary nodule candidates especially with respect to the critical issue of malignant or benign. The classical approach here is to generate some diagnostic features from the appearance of the nodule images and to perform classification from these to determine malignancy. A number of research studies have shown encouraging results; however, these studies have either used non-screening nodules and images, which have a vastly larger size and higher quality or did not separate out the contribution of nodule size, which is highly predictive of malignancy in LCS populations, from the other image features. A recent study [3] has shown that after compensating for size, for LCS CT images, the other image features provide only a moderate amount of additional information. This information is insufficient for a diagnosis by itself but may be used to refine follow up decisions. The measurement of nodule growth rate from two or more CT scans has been shown to be highly predictive of nodule malignancy status [4]. Since at least a second scan is required this method should be considered as a follow up procedure among other clinical follow up methods. The main barrier to clinical implementation of this method is that it requires the computing of the difference of the two CT scans, which is highly dependent on the geometric image quality of each scan. Unfortunately, there exists no agency or process by which this quality is monitored or measured on current scanners and without any scanner calibration imprecise results may occur. Correct use of this method requires careful attention to details. CT scans acquired for LCS also image other critical organs that are at risk for the screening population. Companion diagnostics refers to computer analysis for conditions other than lung cancer from the periodic LCS CT images. Conceptually, this is similar to a blood test where additional conditions may be evaluated from a single patient interaction. Therefore, the automatic risk factor assessment of these additional regions provides additional benefit without requiring additional imaging for the LCS population. Work in this area is still at an early stage. Research targets for automated evaluation reported in the literature include: lung (emphysema and COPD), cardiac (coronary artery calcium, aorta profile and calcium), breast (density assessment), and bone (vertebral body density evaluation). Computer aided methods will inevitably make major contributions to increasing the efficiency and benefit of LCS as they transition from research prototypes to clinical practice. More sophisticated computer algorithms and modern machine learning techniques will greatly improve CAD performance; however, such methods require very large training image datasets. Research studies to date typically involve 100 images examples or less; future algorithm development can greatly benefit by the millions of images that will be acquired with LCS practice. References [1] Enquobahrie A A, Reeves A P, Yankelevitz D F and Henschke C I, “Automated Detection of Small Pulmonary Nodules in Whole Lung CT Scans”, Acad Radiol, 14(5): 579-593, 2007. [2] Zhao Y, de Bock G H, Vliegenthart R, van Klaveren R J, Wang Y, Bogoni L, de Jong P A, Mali W P, van Ooijen P M A and Oudkerk M, “Performance of computer-aided detection of pulmonary nodules in low-dose CT: comparison with double reading by nodule volume”, Eur Radiol, 22(10): 2076-2084, 2012. [3] Reeves A P, Xie Y and Jirapatnakul A, “Automated pulmonary nodule CT image characterization in lung cancer screening”, IJCARS, doi: 10.1007/s11548-015-1245-7, 2015. [4] Reeves A P, “Measurement of Change in Size of Lung Nodules”. In Li Q, Nishikawa R M (ed) Computer-Aided Detection and Diagnosis in Medical Imaging, Taylor & Francis, Chapter 11, 2015.

      Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.