| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
* From the Department of Molecular and Medical Pharmacology (Drs. Schiepers, Weber, Fueger, Phelps, and Czernin), Ahmanson Biological Imaging Center, UCLA David Geffen School of Medicine, Los Angeles, CA; Department of Radiological Sciences (Dr. Crawford), UCLA David Geffen School of Medicine, Los Angeles, CA; and Department of Radiology (Dr. Halpern), Medical Diagnostic Division, Medical University Vienna, Vienna, Austria.
Correspondence to: Johannes Czernin, MD, UCLA School of Medicine, Nuclear Medicine, AR 128 CHS, 10833 Le Conte Ave, Los Angeles, CA 90095-6942; e-mail: jczernin{at}mednet.ucla.edu
 |
Abstract |
|---|
 TOP Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Methods: Thirty-six patients (17 men and 19 women) with NSCLC underwent staging with integrated PET/CT followed by mediastinal lymph node dissection and tumor resection. Twenty-five of the 36 patients (69%) underwent separate CT studies for software fusion of images. Two blinded reviewers analyzed in consensus all PET images, and an experienced radiologist was added to assess integrated and software-fused PET/CT images. Histopathologic findings served as "gold standard" for determining the diagnostic accuracy of all modalities.
Results: Reviewers examining PET and integrated PET/CT classified T stage accurately in 67% (20 of 30 patients) and 97% (29 of 30 patients), respectively (p < 0.05). Overall, interpretations based on PET staged 57% (17 of 30 patients) correctly, overstaged 6 patients (20%), and understaged 7 patients (23%). Interpretations based on integrated PET/CT correctly staged 83% (25 of 30 patients), overstaged 3 patients (10%), and understaged 2 patients (7%). The overall staging accuracy of integrated PET/CT was significantly higher than that of PET (p < 0.05). Automatic software fusion of separately obtained PET and CT studies was successful in 68% of the patients but failed in 32%. In successful software fusion cases, the results of software fusion with regards to T stage and N stage were not different from integrated PET/CT.
Conclusions: Integrated PET/CT compared with PET alone was associated with 26% points-greater overall diagnostic accuracy (p = 0.01). The software fusion method failed to provide acceptable coregistration in > 30% of the patients.
Key Words: CT • dual-modality imaging • fusion imaging • positron emission tomography • non-small cell lung cancer
|
Introduction |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Fluorodeoxyglucose positron emission tomography (FDG-PET) imaging is the most accurate imaging modality for lung cancer staging but faces several major challenges. The first is its limited specificity due to increased glycolytic activity of benign tumors and inflammatory tissue, in addition to that of malignant tumors.3 Secondly, its anatomic resolution is limited, precluding exact localization of glucose alterations to specific anatomic structures. Thirdly, mildly hypermetabolic primary or metastatic lesions might be too small to be identified with FDG-PET. Finally, some malignancies such as pure bronchoalveolar carcinoma may not exhibit any discernible increase in glycolytic activity.
Integrated PET/CT and software fusion of positron emission tomography (PET) and CT images can help to overcome these limitations to a certain degree.4 With integrated PET/CT, anatomic and molecular information can accurately be coregistered.5
Another way to achieve reasonable coregistration between metabolic and anatomic imaging is the software fusion of separately obtained PET and CT images. Software fusion is relatively inexpensive, but to our knowledge, its clinical performance has not been compared to that of integrated PET/CT.
Some studies67 have already compared PET and integrated PET/CT for staging in NSCLC and have demonstrated additional gains in diagnostic accuracy. However, the aim of the current study was to compare the diagnostic accuracy between FDG-PET, integrated PET/CT, and software fusion for the staging of NSCLC.
|
Materials and Methods |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
PET/CT Imaging Protocol
All patients were advised to fast for at least 6 h prior to the integrated PET/CT examination. Sixty minutes prior to the integrated PET/CT scan, all patients received 7.77 megabecquerels per kilogram of fluorodeoxyglucose (FDG). Patients were scanned from the mid-thigh level to the base of the skull in the "arms-up" position. CT studies were performed without IV contrast application. All patients were advised not to speak, chew, or move during the uptake period of FDG and the scan.
Integrated PET/CT studies were acquired with a scanner (Biograph Scanner; Siemens Medical Solutions; Hoffman Estates, IL). This system consists of an ECAT ACCEL PET system (CTI; Knoxville, TN), without septa and transmission sources, and a Somatom Emotion duo radiograph CT system (Siemens Medical Systems; Iselin, NJ).
After determining the imaging field, a 80- to 110-s whole-body CT acquisition was performed using the following parameters: 130 kilovolt peak, 120 mA, 1-s tube rotation, 4-mm slice collimation, 5-mm reconstruction slice thickness, and a table feed of 8 mm per rotation (ie, pitch = 2). All studies were performed at shallow breathing. On completion of the CT portion, the PET emission data were acquired in the three-dimensional mode using a weight-based protocol as described previously.8
PET/CT Image Reconstruction
CT images were reconstructed using conventional-filtered back-projection at 3.4-mm axial intervals to match the slice separation of the PET data. PET images were reconstructed using iterative algorithms (ordered-subset expectation maximization, two iterations, eight subsets) to a final image resolution of 8.8-mm full width half maximum. For attenuation correction, CT Hounsfield units were mapped to the linear attenuation coefficients of 511 keV.59
Software PET and CT Image Fusion
Twenty-five of the 36 patients (69%) underwent separate noncontrast CT scans within 6 weeks of the integrated PET/CT study. No therapeutic interventions were performed between the integrated PET/CT and the separate CT study. CT studies were performed with a spiral CT scanner (HiSpeed Spiral CT/I; GE Medical Systems; Milwaukee, WI) using the following parameters: 130 kilovolt peak, 80 mA, 0.8-s tube rotation, 2.5-mm slice collimation, 3-mm reconstruction slice thickness, and a table feed of 10 mm per rotation (ie, pitch = 2). All studies were performed at breath-holding during maximal inspiration.
Software fusion of PET emission data and CT was performed using a fully automated commercially available software approach (Mirada Solutions; Oxford, UK). This approach is based on similarity measures corresponding to different modeling assumptions, such as correlation coefficient, correlation ratio, and mutual information.10 After loading the PET and CT data, a linear rigid fusion was applied first. The linear algorithm is limited to translation, rotation, and scaling of the images.1112 This was followed by a nonlinear fusion algorithm. This approach, also known as image warping, allows for corrections of configuration changes, as for instance induced by breathing or movement of internal organs.12 No manual alignment of images was used since this approach is known to be unreliable because of the adjustment of multiple parameters.
To determine the degree of misalignment between PET and CT both for hardware and software fusion, the center of hypermetabolic lesions (ie, lung tumors or lymph nodes) was depicted as landmarks on CT or PET images. Successful software fusion was defined as coregistration of the hypermetabolic region on PET to a distance of < 2 cm in the x, y, and z directions of the anatomically localized lesion on CT. The distance of 2 cm was chosen because this represents two times the size of lymph nodes deemed positive by CT and approximately two times the PET scanner resolution after image reconstruction. Therefore, a misalignment of 2 cm was considered to be of clinical relevance for successful software fusion. PET and CT foci were then measured in the x, y, and z directions, and the overall distance (D) between landmarks on PET and CT was calculated using the following formula:
 |
x, y, and z are the distances between the corresponding landmarks in the x, y, and z directions.
Image Analysis
Based on imaging findings, tumor stage was assessed using the most recent American Joint Committee on Cancer13 TNM classification for NSCLC. First, PET images were interpreted by consensus by two readers who were blinded for the CT portion of the study. They were also blinded for any clinical information other than the diagnosis of lung cancer. A radiologist was subsequently added to interpret integrated and software-fused PET/CT images. We allowed for sufficient time between the evaluation of integrated and software-fused PET/CT images (ie, 1 month) to avoid the recognition of certain patterns by the radiologist, thereby introducing a bias in the software fusion interpretation. The consensus approach was chosen to avoid confounding the data by interobserver variability.
For PET interpretation, primary tumors as well as other lesions were classified as either positive or negative for malignancy. Lesions were considered positive if they exhibited focally increased FDG uptake above the normal background activity (for instance lung or mediastinum).
For integrated and software-fused PET/CT, primary tumors were considered positive for malignancy if FDG uptake was increased or if the tumor had a "malignant appearance" on CT. Lymph nodes of any size were only considered positive if FDG uptake was increased above background. This approach was chosen because PET lymph node staging of the mediastinum is significantly more accurate than CT staging.14 T, N, and M stages were assigned for each modality for each patient.
Statistical Analysis
For paired comparisons (sensitivity, specificity, and accuracy) between PET and integrated and software-fused PET/CT, the McNemar test was used for calculation. Unpaired comparisons (positive predictive value [PPV] and negative predictive value [NPV]) between PET and integrated and software-fused PET/CT were evaluated by using the 
2 test. Sensitivity, specificity, PPV, and NPV were only calculated for N stage but not for T stage and overall TNM stage. All values were assessed on a patient-by-patient analysis. Integrated and software-fused PET/CT accuracies were determined for both devices in the 17 patients for whom successful software fusion was available. All of these 17 patients underwent tumor resection and therefore had complete TNM stage assessment. Confidence intervals (CIs) were calculated using standard methods. Pathology findings served as the "gold standard" for T stage, N stage (lymph node stations), and M stage.
|
Results |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
T Stage
The pathologic T stages available in the 30 patients who underwent surgery are as follows: T1 (n = 16; 54%), T2 (n = 9; 30%), T3 (n = 4; 13%), and T4 (n = 1; 3%). Mean ± SD size of the excised tumors was 3.0 ± 2.1 cm (range, 1.0 to 10.0 cm). PET alone classified the T stage correctly in 20 patients (67%; CI, 47 to 83%); 9 patients were understaged (30%) and 1 patient was overstaged (3%; Table 1
). Reasons for understaging included three FDG-negative primary tumors (bronchoalveolar carcinoma, n = 2; adenocarcinoma with bronchoalveolar features, n = 1) [Fig 1
], undetected pleural or chest wall infiltration (n = 3) [Fig 2
], and underestimating tumor size (n = 3). In the one overstaged patient, pleural infiltration was predicted by PET but was not confirmed by surgery.
|
|
|
N Stage
Twenty-six of the 36 patients (72%) had N0 disease, 9 patients (25%) had N2 disease, and 1 patient (3%) had N3 disease. Using PET alone, reviewers correctly assigned the N stage in 25 of 36 patients (69%; CI, 52 to 84%) [Table 1]. Overstaging occurred in six patients (17%). Two of these patients were upstaged from N0 to N1, two patients were upstaged from N0 to N2, and two patients were upstaged from N0 to N3. Overstaging was due to granuloma disease in three of the six patients and was due to misinterpretation of vascular pool as lymph node involvement in another two patients (Fig 3
). The remaining patient had histologic evidence of anthracosis. Five patients were understaged (14%): three patients from N2 to N0 and two patients from N2 to N1. Understaging was explained by subcentimeter lymph nodes without increased FDG uptake in three patients. The remaining two patients were incorrectly assigned to a lower nodal stage.
|
In clinical practice, the treatment approach of NSCLC does not differ between N0 and N1 disease. Thus, a subset analysis was performed, whereby N0 and N1 disease were lumped together and the accuracy of PET was compared to that of integrated PET/CT. This revealed that the PET accuracy increased to 75% (CI, 58 to 88%), while the accuracy of integrated PET/CT remained unchanged at 78% (CI, 61 to 90%; p = NS). The sensitivity, specificity, PPV, NPV, and accuracy of PET and integrated PET/CT regarding lymph node staging are listed in Table 2 .
|
Overall TNM Stage
The stage distribution in the 30 patients in whom complete TNM stage was available was as follows: stage IA (n = 15, 50%), stage IB (n = 7, 24%), stage IIB (n = 3, 10%), stage IIIA (n = 4, 13%), and stage IIIB (n = 1, 3%). PET and integrated PET/CT results for TNM stage are shown in Table 1. TNM stage was correctly assessed with PET in 17 of 30 patients (57%; CI, 37 to 75%) and with integrated PET/CT in 25 of 30 patients (83%; CI, 65 to 94%; p = 0.01 vs PET).
Both PET and integrated PET/CT assigned an incorrect TNM stage in 5 of 30 patients (17%). Three of these patients were overstaged due to bilateral granuloma disease. Two patients with N2 disease were understaged by both modalities. Integrated PET/CT correctly determined the TNM stage in eight patients in whom the PET-derived stage was incorrect.
Software Fusion
Software fusion was successful in 17 of the 25 patients (68%) in whom fusion was attempted but failed in the remaining 8 patients (32%). Integrated PET/CT and software fusion arrived at identical TNM stages in the 17 patients in whom fusion was successful. However, the degree of misalignment was consistently higher for software than hardware fusion (Fig 4
). The distances between the center of the hypermetabolic areas and the anatomic correlates on the CT images averaged 7.2 ± 3.3 mm3 for integrated PET/CT and 15.4 ± 8.3 mm3 for software-fused PET/CT (p < 0.05).
|
|
Discussion |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
For the current analysis, we describe in detail the impact of PET/CT on T stage and N stage. This approach was chosen to facilitate a comparison between the current and previous studies67 that also investigated lung cancer staging with PET/CT. For this reason, we did not use the well-established international system for staging lung cancer.15
As expected and as previously reported, integrated PET/CT assigned the T stage with a higher accuracy than PET.7 The current accuracy for T staging with integrated PET/CT (97%) is consistent with the data published by Lardinois et al7 (98%) and Antoch et al6 (94%).
The advantage of integrated PET/CT with regards to T staging was solely derived from the CT information. This was expected since PET cannot be used to measure tumor size and thus cannot be used to reliably determine T stage. The spatial resolution of PET is limited, which precludes precise size determination. The mean diameter of primary tumors was 3.0 ± 2.1 cm in the current study. Thus, the average tumor size was close to the threshold of 3 cm that separates T1 stage from T2 stage. Secondly, CT uncovered tumor infiltration into adjacent structures such as pleura, mediastinum, or chest wall, a prerequisite for differentiating T2 disease from T3 and T4 disease (Fig 2). Finally, CT added important information in tumors without increased glycolytic activity (Fig 1). Such tumors include, among others, pure bronchoalveolar carcinoma or mucine-producing adenocarcinoma.16 Despite absent glucose metabolic activity, the CT findings were sufficiently suspicious to interpret the integrated PET/CT findings as positive in three patients. It is, however, interesting to note that both Lardinois et al7 and Antoch et al6 reported superior T staging with integrated PET/CT than with CT alone. The reasons for these observations are unclear, since no specific advantages of PET for T staging were mentioned in their studies.
Both PET and integrated PET/CT had limited accuracies for lymph node assessment of 69% (CI, 52 to 84%) and 78% (CI, 61 to 90%), respectively (p = NS). However, integrated PET/CT correctly changed the N stage in 3 of 36 patients (8%). In two of these patients, PET readers misinterpreted physiologic blood pool activity as nodal disease (Fig 3). In the remaining patient, the nodal station was incorrectly assigned by PET but correctly determined by integrated PET/CT.
The accuracy for mediastinal lymph node staging with CIs ranging from 58 to 88% for PET and 61 to 90% for integrated PET/CT is lower than previously reported.6714171819 The lower-than-expected sensitivity is explained by the small number of patients with lymph node involvement. In addition, 22 patients with stage I and thus early disease were included who had subcentimeter yet positive lymph nodes that were missed by PET. The specificity of FDG-PET and integrated PET/CT was lower than previously reported because of the high incidence of granuloma disease in southern California. Lardinois et al7 and Antoch et al6 reported that PET and integrated PET/CT staged lymph node involvement with accuracies of 87% and 89%, respectively. Marginally improved lymph node staging by integrated PET/CT was reported by Antoch et al6 but not by Lardinois et al.7 However, neither of these studies provided detailed data with regards to the exact pathologic nodal stage or the number of patients with lymph node involvement. Thus, a comparison between the current and their studies is not feasible. The current study confirms, however, the findings of Lardinois et al7 and Antoch et al6 that integrated PET/CT offered no significant benefit for nodal staging when compared to PET alone.
Correct assessment of the N stage remains a challenging issue for both PET and integrated PET/CT. Benign diseases such as sarcoidosis, tuberculosis, and fungal infection can result in false-positive PET findings. Obviously, false-positive lymph nodes due to benign inflammatory diseases are also seen on integrated PET/CT. However, CT characterization of lymph nodes as malignant or benign by size criteria is inadequate. For instance, nodes measuring < 1 cm in diameter have a likelihood of 3 to 16% of harboring cancer, while large lymph nodes can be reactive and benign.20212223 As expected, the reasons for false-positive PET and integrated PET/CT findings in the current study included inflammatory diseases, and false-negative findings were largely explained by small-sized lymph nodes. The small gain in accuracy for lymph node staging by integrated PET/CT was due to appropriate assignment of focally increased FDG uptake to blood pool rather than lymph node activity. This emphasizes the importance of anatomic information in conjunction with PET imaging for appropriate PET image interpretation.
At our institution, mediastinoscopy is still performed in most lung cancer patients for N stage assessment, even if integrated PET/CT reveals no evidence of mediastinal lymph node involvement. However, integrated PET/CT findings are frequently used to guide mediastinoscopy to areas of unexpected lymph node involvement. This algorithm might change with the availability of more extensive literature addressing the accuracy of FDG-PET/CT for mediastinal lymph node staging.
Since the study population was biased toward likely surgical candidates, the relative accuracies of integrated PET/CT and PET with regards to M stage could not be determined.172425 Nevertheless, integrated PET/CT and PET identified two sites of metastatic disease in two patients.
One of the potential advantages of integrated PET/CT might be the assessment of distant metastases, ie, the M stage. For example, bone metastases are often indistinguishable from degenerative bony processes by PET alone. Further, sclerotic bone metastases, typical in appearance by CT, are frequently negative on FDG-PET. These issues could, however, not be addressed in the current study since the study design did not allow for a high number of patients who had distant disease. Future studies will be needed to address specific advantages of PET/CT with regards to M staging.
One important aspect of this study was the performance evaluation of a commercially available software fusion package. This uses an intensity-based technique rather than segmenting homologous geometric landmarks, such as surfaces or volumes in the images used.10 The basic principle is to search, in a certain space of transformations, for an intensity similarity between corresponding voxels. We employed a stepwise approach to software fusion whereby rigid, linear fusion was applied first that was followed by deformable fusion also termed image warping. Software fusion is frequently marketed as an inexpensive "push button" alternative for hardware fusion. This is, as demonstrated by our data, not quite as simple and straightforward. Despite using two processing steps and a liberal definition of misregistration of 2 cm, software fusion was only successful in 68% of the patients. The argument could be made that even in patients with fusion failure, one could have arrived at the correct TNM stage by, for instance, reviewing PET and CT images "side by side." This approach, however, would defeat the purpose of the software approach, which aims at fusing with a high accuracy individual lesions to accommodate for such diverse applications as radiation or biopsy planning. The 68% success rate is consistent with a study12 that emphasized the need to include transmission PET scans to improve the performance of software fusion; using this approach, these researchers increased the success rate of software fusion from approximately 70 to 95% of the patients. Even in patients with successful software fusion, the alignment between PET and CT images was significantly less accurate (p < 0.05) than for integrated PET/CT (Fig 4). This is again not surprising, since CT studies obtained at different sites and different times are not optimized for fusion with PET.
As a bias against software fusion, the breathing protocols differed between the CT and the PET acquisition. This likely accounted for the higher degree of misregistration by software than by hardware fusion. However, the findings do reflect a "real-life" clinical scenario, whereby PET and CT images are almost always acquired under very different conditions and protocols. Standardization of PET and CT with regards to breathing protocols will be difficult if not impossible to implement within and across institutions. Thus, differences in patient position and respiratory status likely account for this finding. Peripheral lung lesions tended to be more prone to misalignment than central lung lesions or mediastinal lymph nodes. Respiratory lung motion that is more prominent in the lung periphery might explain this observation. This is supported by the fact that the misalignment was highest in the z-axis, for which the most pronounced variability due to lung motion would be expected.
Several conclusions can be drawn from the relatively high failure rate of software fusion. First, transmission scans need to be included in the fusion process to increase success rates as demonstrated by Slomka et al.12 Secondly, future studies need to determine the ability of software fusion to improve the registration of hardware-fused images. Thirdly, other applications of software fusion such as the comparison of serial PET or integrated PET/CT studies should be explored. Fourthly, the small sample size precludes a final assessment of the clinical validity of this approach. However, the study suggests considerable limitations in the clinical usefulness of software fusion if PET and CT images are not acquired under near identical conditions. Finally, users of fusion software need to understand the limitations of this approach.
As a limitation of the study, the readers were aware that all patients had NSCLC. Thus, in the three patients with primary tumors without increased glycolytic activity, the radiologist’s interpretation might have been biased toward malignancy. It should be mentioned, however, that the primary CT mass was considered highly suspicious for malignancy by the reader despite absence of glucose metabolic activity.
Another limitation of the study is related to the use of mediastinoscopy as a "gold standard." Dillemans et al26 reported in a study of 569 patients a sensitivity, specificity, and accuracy for mediastinoscopy of 72%, 100%, and 89%, respectively. Thus, a considerable number of abnormal lymph nodes are missed by mediastinoscopy. However, we are unaware of a more reliable "gold standard" than mediastinoscopy and thoracotomy for comparison with imaging findings.
In conclusion, integrated PET/CT improves overall TNM staging when compared to PET alone. The advantage is derived from improved T staging and to a small degree also from improved N staging. Software fusion of separately obtained PET and CT images failed in > 30% of the patients but arrived at the same results as integrated PET/CT in the remaining patients.
|
Footnotes |
|---|
Received for publication February 6, 2005. Accepted for publication March 15, 2005.
|
References |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  D. Lardinois New Horizons in Staging for Non-Small-Cell Lung Cancer J. Clin. Oncol., April 20, 2006; 24(12): 1785 - 1787. [Full Text] [PDF]
|
|
|
|
 |
|
 T. Bunyaviroch and R. E. Coleman PET Evaluation of Lung Cancer J. Nucl. Med., March 1, 2006; 47(3): 451 - 469. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
