Volume of lung nodules is an important biomarker, quantifiable from computed tomography (CT) images. The usefulness of volume quantification, however, depends on the precision of quantification. Experimental assessment of precision is time consuming. A mathematical estimability model was used to assess the quantification precision of CT nodule volumetry in terms of an index (e′), incorporating image noise and resolution, nodule properties, and segmentation software. The noise and resolution were characterized in terms of noise power spectrum and task transfer function. The nodule properties and segmentation algorithm were modeled in terms of a task function and a template function, respectively. The e′ values were benchmarked against experimentally acquired precision values from an anthropomorphic chest phantom across 54 acquisition protocols, 2 nodule sizes, and 2 volume segmentation softwares. e′ exhibited correlation with experimental precision across nodule sizes and acquisition protocols but dependence on segmentation software. Compared to the assessment of empirical precision, which required ∼300 h to perform the segmentation, the e′ method required ∼3 h from data collection to mathematical computation. A mathematical modeling of volume quantification provides efficient prediction of quantitative performance. It establishes a method to verify quantitative compliance and to optimize clinical protocols for chest CT volumetry.
The purpose of this study was to establish volumetric exchangeability between real and computational lung lesions in CT. We compared the overall relative volume estimation performance of segmentation tools when used to measure real lesions in actual patient CT images and computational lesions virtually inserted into the same patient images (i.e., hybrid datasets). Pathologically confirmed malignancies from 30 thoracic patient cases from Reference Image Database to Evaluate Therapy Response (RIDER) were modeled and used as the basis for the comparison. Lesions included isolated nodules as well as those attached to the pleura or other lung structures. Patient images were acquired using a 16 detector row or 64 detector row CT scanner (Lightspeed 16 or VCT; GE Healthcare). Scans were acquired using standard chest protocols during a single breath-hold. Virtual 3D lesion models based on real lesions were developed in Duke Lesion Tool (Duke University), and inserted using a validated image-domain insertion program. Nodule volumes were estimated using multiple commercial segmentation tools (iNtuition, TeraRecon, Inc., Syngo.via, Siemens Healthcare, and IntelliSpace, Philips Healthcare). Consensus based volume comparison showed consistent trends in volume measurement between real and virtual lesions across all software. The average percent bias (± standard error) shows -9.2±3.2% for real lesions versus -6.7±1.2% for virtual lesions with tool A, 3.9±2.5% and 5.0±0.9% for tool B, and 5.3±2.3% and 1.8±0.8% for tool C, respectively. Virtual lesion volumes were statistically similar to those of real lesions (< 4% difference) with p >.05 in most cases. Results suggest that hybrid datasets had similar inter-algorithm variability compared to real datasets.
This study aimed to develop and compare two methods of inserting computerized virtual lesions into CT datasets. 24
physical (synthetic) nodules of three sizes and four morphologies were inserted into an anthropomorphic chest phantom
(LUNGMAN, KYOTO KAGAKU). The phantom was scanned (Somatom Definition Flash, Siemens Healthcare) with and
without nodules present, and images were reconstructed with filtered back projection and iterative reconstruction
(SAFIRE) at 0.6 mm slice thickness using a standard thoracic CT protocol at multiple dose settings. Virtual 3D CAD
models based on the physical nodules were virtually inserted (accounting for the system MTF) into the nodule-free CT
data using two techniques. These techniques include projection-based and image-based insertion. Nodule volumes were
estimated using a commercial segmentation tool (iNtuition, TeraRecon, Inc.). Differences were tested using paired t-tests
and R2 goodness of fit between the virtually and physically inserted nodules. Both insertion techniques resulted in nodule
volumes very similar to the real nodules (<3% difference) and in most cases the differences were not statistically
significant. Also, R2 values were all <0.97 for both insertion techniques. These data imply that these techniques can
confidently be used as a means of inserting virtual nodules in CT datasets. These techniques can be instrumental in building
hybrid CT datasets composed of patient images with virtually inserted nodules.
The purpose of this study was to develop a 3D quantification technique to assess the impact of imaging system on depiction of lesion morphology. Regional Hausdorff Distance (RHD) was computed from two 3D volumes: virtual mesh models of synthetic nodules or “virtual nodules” and CT images of physical nodules or “physical nodules”. The method can be described in following steps. First, the synthetic nodule was inserted into anthropomorphic Kyoto thorax phantom and scanned in a Siemens scanner (Flash). Then, nodule was segmented from the image. Second, in order to match the orientation of the nodule, the digital models of the “virtual” and “physical” nodules were both geometrically translated to the origin. Then, the “physical” was gradually rotated at incremental 10 degrees. Third, the Hausdorff Distance was calculated from each pair of “virtual” and “physical” nodules. The minimum HD value represented the most matching pair. Finally, the 3D RHD map and the distribution of RHD were computed for the matched pair. The technique was scalarized using the FWHM of the RHD distribution. The analysis was conducted for various shapes (spherical, lobular, elliptical, and speculated) of nodules. The calculated FWHM values of RHD distribution for the 8-mm spherical, lobular, elliptical, and speculated “virtual” and “physical” nodules were 0.23, 0.42, 0.33, and 0.49, respectively.