AJR InPractice
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
 QUICK SEARCH:   [advanced]


     


This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Rosen, M. P.
Right arrow Articles by Levine, D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Rosen, M. P.
Right arrow Articles by Levine, D.
Social Bookmarking
 Add to CiteULike   Add to Complore   Add to Connotea   Add to Del.icio.us   Add to Digg   Add to Reddit   Add to Technorati  
What's this?
AJR 2001; 176:855-860
© American Roentgen Ray Society


Remote Sonographic Interpretation Using a Laser Printer Network

System Performance and Diagnostic Accuracy in Actual Clinical Practice

Max P. Rosen1, Tejas S. Mehta, Rebecca Bromberg, Sandra L. Kelly and Deborah Levine

1 All authors: Department of Radiology, Beth Israel Deaconess Medical Center and Harvard Medical School, 330 Brookline Ave., Boston, MA 02215.

Received June 14, 2000; accepted after revision September 15, 2000.

 
Presented at the annual meeting of the Radiological Society of North America, Chicago, November 1999.

Address correspondence to M. P. Rosen.


Abstract
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 
OBJECTIVE. The purpose of our study was to evaluate the technical and clinical performance of remote sonographic interpretation using a laser printer network connecting a community-based imaging center and an academic medical center.

SUBJECTS AND METHODS. During a 3-month period, 161 consecutive sonographic examinations were performed at a community-based imaging center and all 161 patients were enrolled in the study. Seventy-one (44%) of 161 examinations were interpreted on-site at the community-based imaging center, and 90 (56%) of 161 were transmitted over a T-1 line to an academic medical center where the static images were interpreted remotely. For the purposes of this study, the examination time was defined as the interval from the time the technologist started to scan the patient to the time the patient was dismissed from the radiology department. Examination times were recorded for each patient. Follow-up was available for 92 (57%) of 161 studies. Sensitivity and specificity for studies interpreted at the community-based imaging center and at the academic medical center were calculated.

RESULTS. The mean examination time for pelvic sonographic examinations interpreted at the academic medical center (43 min) was significantly longer than for scans interpreted at the community-based imaging center (31 min) (p < 0.01). However, no significant difference was noted in the examination time for abdominal sonography. For all examinations interpreted on-site at the community-based imaging center for which follow-up was available, the sensitivity and specificity were 95% and 100%, respectively. For all examinations interpreted remotely at the academic medical center for which follow-up was available, the sensitivity and specificity were 93% and 90%, respectively. No significant difference was seen in the sensitivity (p = 1.00) or specificity (p = 0.24) of studies interpreted on-site versus remotely.

CONCLUSION. Static sonographic images can be interpreted remotely without loss of sensitivity, but with decreased specificity. However, more time must be allotted for performing pelvic sonography when these examinations are to be interpreted remotely.


Introduction
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 
The current pressures on health care have expanded the scope of outpatient care. Community outpatient clinics are increasingly required to provide greater expertise across a broader spectrum of specialties, including radiology. Academic radiology departments are often asked to provide subspecialty expertise to community-based clinics; the volume of radiology examinations at many of these community clinics is not large enough to support an on-site radiologist.

One possible solution to this dilemma is the implementation of teleradiology, in which images can be digitally transferred, allowing radiologists at an academic medical center to view images acquired at a community-based imaging center almost instantaneously [1]. For static imaging modalities such as CT and MR imaging, essentially no information is lost in transfer. However, electronic transfer of static sonographic images is more complex because potentially helpful information is lost when the radiologist is unable to perform real-time scanning.

When required to interpret remote sonographic images, radiologists at academic medical centers often record the sonographic examination on videotape to avoid losing dynamic information. The videotapes are then delivered to the academic medical center, where they are reviewed at the end of the workday. However, such "batch" interpretation does not allow the academic medical center radiologist to obtain additional images of ambiguous areas before the patient leaves the community-based imaging center. This problem may ultimately result in many patients being called back, which might have been unnecessary had the examinations been reviewed at the time they were conducted.

The purpose of our study was to evaluate the technical and clinical performance of remote sonography interpretation. Specifically, we wanted to document the accuracy of interpretation of static sonographic images using a laser printer network connecting the aforementioned community-based imaging center and academic medical center via a T-1 data link. In addition, we documented the potential impact that our system of remote sonographic interpretation might have on reducing the callback rate for sonographic examinations.


Subjects and Methods
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 
Between July 21, 1998, and October 13, 1998, 161 consecutive sonographic examinations were performed at a community-based imaging center affiliated with an academic medical center. The same sonographer, who has 21 years of sonography experience, performed all examinations. The community-based imaging center is staffed 20 hr per week by one of five board-certified (on-site) general radiologists. Two of these five general radiologists did not have any fellowship training. Two of the five had fellowship training in vascular and interventional radiology, and one of the five had fellowship training in women's imaging. These five radiologists work at the community-based imaging center on a rotating basis. Scans obtained when the on-site radiologist is present are interpreted by the on-site radiologist, who has the opportunity to scan the patient and to talk directly with the patient.

During the remaining 20 hr of each week, the community-based imaging center is not staffed by a radiologist. All sonographic images are transmitted to the academic medical center via a highspeed (T-1) line. At the academic medical center the images are printed on a laser printer (Eastman Kodak, Rochester, NY) and interpreted by one of four board-certified (remote) radiologists. Unlike the community-based imaging center radiologists, the remote radiologists at the academic medical center have either received fellowship training in sonography (n = 3) or had extensive (20 yr) experience in sonography (n = 1).

The academic medical center radiologist has the opportunity to review transmitted static images and ask for additional images before the patient leaves the community-based imaging center. In addition, the technologist and radiologist at the academic medical center discuss each case via telephone, at which time the technologist provides the radiologist with a preliminary report of the findings. An attempt was made to schedule obstetric sonographic examinations when the radiologist was present at the community-based imaging center, and all highrisk obstetric patients were scanned at the academic medical center. Otherwise, no bias existed in scheduling patients at the community-based imaging center. Specifically, we did not attempt to schedule cases that may have been perceived as more difficult or complex when the academic medical center interpretation was available. The same scanning protocols were followed whether the sonographic studies were interpreted by a radiologist at the community-based imaging center or at the academic medical center. The general radiologist at the community-based imaging center interpreted all studies and did not transmit any images to the academic medical center for a second opinion.

Each examination was placed in one of the following categories: abdomen, pelvis, combined abdomen and pelvis, obstetric, small parts (thyroid, scrotal), or vascular. When the sonography was performed, the following information was prospectively obtained and entered into a spreadsheet (Excel; Microsoft, Seattle, WA): the time required to perform each examination (from the time the technologist started to scan the patient to the time the patient left the radiology department), patient age and sex, the location of the radiologist (community-based imaging center or academic medical center), and whether additional views were requested by the radiologist. The radiologists were not aware that the study was being conducted. The study was approved by our hospital's investigation review board. Examination time did not include interpretation time.

At least 10 months after the initial data collection (starting in June 1999), each patient's medical record was reviewed, and clinical, pathologic, or imaging follow-up relevant to the indication for the initial sonography was obtained and recorded. A radiologist who is fellowship-trained in sonography reviewed each sonographic report and graded each sonographic study as showing normal or abnormal findings. The radiologist then compared the sonography report with the record of the patient's follow-up. For each patient for whom follow-up was available, the radiologist categorized the sonography report and follow-up as either concordant or discordant. (The radiologist was unaware of the site at which the study was interpreted and which radiologist interpreted the study.) When discordance existed between the initial sonographic study and the follow-up, the static images of the initial sonographic study and the follow-up study were reviewed to determine which was correct. The sensitivity and specificity for sonography interpreted by the on-site and remote radiologists were then calculated, both with and without correcting for verification bias. The difference between the sensitivity and specificity for on-site and remote interpretation, and the difference in the rate of verification for normal and abnormal findings, were calculated using a two-sided Fisher's exact test.

Differences in examination time and in requests for additional views were calculated for abdominal and pelvic sonographic examinations using the Student's t test and the Fisher's exact test, respectively. Differences in examination time and requests for additional images were not compared for other types of examinations because of the small numbers of patients in each of the other examination categories. For the purpose of this study, "sensitivity" identifies the frequency of sonography reports with abnormal findings supported by abnormalities of the clinical, pathologic, or imaging follow-up. "Specificity" is defined as the frequency of sonography reports with normal findings in the absence of any abnormalities in clinical, pathologic, or imaging follow-up.


Results
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 
One hundred sixty-one consecutive patients were enrolled in this study. Twenty-eight (17%) were men, and 133 (83%) were women. The sonographic studies of 71 patients were interpreted by the on-site radiologist at the community-based imaging center, and the sonographic studies of 90 patients were interpreted by the remote radiologist at the academic medical center. The number, type, and length of time required for each study are presented in Table 1. The length of time a patient was in the radiology department for pelvic sonography was significantly longer when the images were interpreted by the remote radiologist than when they were interpreted by the on-site radiologist (p = 0.01, Student's t test). Patients undergoing pelvic sonography interpreted on-site at the community-based imaging center were more likely (9/11, 82%) to have additional images requested than were patients undergoing pelvic sonography interpreted remotely at the academic medical center (21/37, 57%), although the difference was not significant, (p = 0.17, Fisher's exact test). Overall, additional images were requested in 13 (18%) of 71 cases interpreted on-site at the community-based imaging center versus 27 (30%) of 90 cases interpreted remotely at the academic medical center (p = 0.10, Fisher's exact test).


View this table:
[in this window]
[in a new window]

 
TABLE 1 Number, Type, and Time Required for 161 Sonographic Studies Interpreted by On-Site and Remote Radiologists

 

Twenty-five (35%) of 71 sonographic studies interpreted on-site showed abnormal findings compared with 26 (29%) of 90 studies interpreted remotely (Tables 2 and 3). Follow-up was available for 92 (57%) of 161 patients, and results were determined by clinical follow-up (n = 24), additional imaging studies (n = 37), or pathology (n = 31). Follow-up was available for 47 (66%) of 71 patients whose sonograms were interpreted on-site and for 45 (50%) of 90 patients whose sonograms were interpreted remotely. The mean length of time for follow-up was 91 ± 90 days (range, 1-375 days) for studies interpreted on-site and 95 ± 101 days (range, 1-442 days) for studies interpreted remotely. For studies in which follow-up was available, the sensitivity and specificity for studies interpreted on-site were 95% and 100%, respectively; and the sensitivity and specificity for studies interpreted remotely were 93% and 90%, respectively. No significant difference was noted in the sensitivity (p = 1.00) or specificity (p = 0.24) of studies interpreted on-site versus those interpreted remotely. Our study (n = 92 patients for whom follow-up was available) had 80% power to detect a 20% difference in sensitivity or specificity, using a two-sided test, or 71% power to detect a 20% difference in sensitivity or specificity using a one-sided test.


View this table:
[in this window]
[in a new window]

 
TABLE 2 Sonographic Interpretations Performed by On-Site Radiologists

 

View this table:
[in this window]
[in a new window]

 
TABLE 3 Sonographic Interpretations Performed by Remote Radiologists

 

The probability of follow-up for sonography having abnormal findings was 80% (20/25) for studies interpreted on-site and 62% (16/26) for studies obtained remotely. The probability of follow-up for sonography having normal findings was 59% (27/46) for studies interpreted on-site and 45% (29/64) for studies obtained remotely. The probability of follow-up was not significantly different (p = 0.22 [abnormal findings], p = 0.18 [normal findings] for studies interpreted on-site or remotely). When corrected for verification bias [2], the sensitivity and specificity for sonographic studies interpreted on-site were 94% and 100%, respectively. The sensitivity and specificity for sonographic studies interpreted remotely were 91% and 93%, respectively. No significant difference was seen in sensitivity (p = 0.19) or specificity (p = 0.15) for studies interpreted on-site or remotely.

Among the sonographic studies interpreted by the on-site radiologist was one false-negative study. That study involved a 44-year-old woman who presented with nonspecific abdominal pain. Sonography showed a fatty liver and gallstones but a normal-appearing spleen on a single sagittal view. Seven weeks later the patient underwent contrast-enhanced abdominal CT to evaluate persistent abdominal pain. CT showed two wedge-shaped areas of low attenuation in the spleen that were thought to be caused by splenic infarcts. Although it was possible that the splenic infarcts occurred in the interval between examinations, this finding was categorized as false-negative. No false-positive studies were among those interpreted by the on-site radiologist.

Among the sonographic studies interpreted remotely were three false-positive studies. One case was an initial sonographic examination in which an endometrial polyp was questioned by the remote radiologist, but follow-up sonographic examinations showed a normal endometrium. Review of the static images from the initial sonography indicated that the remote interpretation of this study was likely an "overcall" resulting from slightly suboptimal static images of the endometrium.

The second case involved a patient with a fibroid uterus in whom a single sagittal image of each kidney was obtained to assess hydronephrosis. Because one image showed a questionable renal lesion (Fig. 1A), the remote radiologist recommended renal sonography with multiple sagittal and transverse images. The questionable lesion was not present on follow-up renal sonography (Fig. 1B) in which multiple views of the kidney were obtained.



View larger version (92K):
[in this window]
[in a new window]
[as a PowerPoint slide]
 
Fig. 1A. 47-year-old woman with pelvic pain. Sagittal sonogram of right kidney acquired at community-based imaging center shows hypoechoic lesion (arrow) in middle third of kidney.

 


View larger version (172K):
[in this window]
[in a new window]
[as a PowerPoint slide]
 
Fig. 1B. 47-year-old woman with pelvic pain. Second sonogram, which was obtained at academic medical center, fails to replicate lesion.

 

The third case consisted of scrotal sonography in which the remote radiologist could not completely clear the echotexture of the testes from the static images because of a suggestion of a hypoechoic focus in the testes (Fig. 2). Follow-up sonography, which was recommended and performed 3 months later with the radiologist present (and able to scan the patient in real-time), showed normal testes bilaterally. Although it is possible that an acute process such as an infection resolved in the time between examinations, we considered these findings to be false-positive for the purposes of our study.



View larger version (164K):
[in this window]
[in a new window]
[as a PowerPoint slide]
 
Fig. 2. 21-year-old-man with right testicular pain. Transverse sonogram of testes acquired at community-based imaging center shows focal hypoechoic lesion (arrow) that was not replicated on real-time scanning by radiologist at academic medical center 3 months later (not shown).

 

Among the sonographic studies interpreted remotely was one false-negative case in which pelvic sonography was initially interpreted as showing a thickened endometrium and normal adnexa. Subsequent hysterosalpingography showed a normal endometrial cavity and bilateral hydrosalpinx. This case was classified as false-negative (because of the missed hydrosalpinx) and not false-positive (because of the thickened endometrium) because we were unable to locate the static images for review. In addition, the presence of a normal endometrial cavity on the hysterosalpingogram does not preclude the presence of thickened endometrium from other causes.


Discussion
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 
Changes in health care have resulted in a shift in emphasis from inpatient to ambulatory care. The economic pressures associated with these changes have resulted in the increased use of teleradiology, which enables tertiary care centers to provide care to community-based centers. This use of teleradiology is easily done with modalities such as CT or MR imaging that produce static images because no information is lost in the transfer of the static images to a remote site.

Interpretation of sonography from a remote site is much more challenging because sonographic image acquisition is operator-dependent. Studies have shown that an active role of the radiologist is essential for accurate reporting of sonographic images [3, 4]. Because of this unique characteristic of sonography, some practices have invested in remote interpretation of sonography via various modes of video transmission [5]. Their results have been promising. However, a sophisticated video setup not only requires additional financing but also demands more of the radiologists' time [6]. The purpose of our study was to evaluate the technical and clinical performance of remote sonographic interpretation using a laser printer network connecting a community-based imaging center to our main academic medical center via a T-1 data link. This system was already in place between our two sites and so required no additional cost.

Of the 47 sonograms interpreted by the on-site radiologist that had follow-up, 46 interpretations (98%) were concordant with the follow-up findings. The remaining one was a case of a normal-appearing spleen seen on sonography. Follow-up CT performed 8 weeks later showed splenic infarcts. Possible explanations for this discrepancy are that the splenic infarcts were not visible at the time of the sonographic study; the lesions were missed by the on-site radiologist, resulting in a false-negative study; or the embolic events occurred in the time between sonography and CT. No cases existed in which an abnormality was reported by an on-site radiologist but was found to be normal (false-positive) at follow-up.

Of the 44 sonograms interpreted by the remote radiologist that had follow-up, 40 interpretations (91%) were concordant with the follow-up findings and four (9%) were discordant. One of these discordant cases involved evaluation of the endometrium. The remote radiologist thought that an endometrial polyp was present, but this was not confirmed on a follow-up examination at which a second radiologist had the opportunity to scan the patient herself. This overcall can likely be attributed to the added visual information about the endometrium provided by real-time scanning.

In the case of the missed hydrosalpinx, it is possible for a hydrosalpinx to be visualized on hysterosalpingography but not on sonography if the tube is obstructed but decompresses by draining into the uterus. In this situation, the obstructed tube may not be fluid-filled and thus may not be visible on sonography. If this was the case, the "missed" hydrosalpinx may have been because of the limitations of sonography rather than an error by the radiologist.

In the case of a questionable renal cyst seen at the time of pelvic sonography, the patient returned for complete renal sonography, which would not have been necessary had the radiologist been able to scan the patient at the time of the initial pelvic sonography. This case represented one of two call-backs in our series. Because additional views were requested and performed in 27 (30%) of 90 studies interpreted remotely and two patients needed to be called back for additional imaging, we can assume that the net impact of our system (compared with batch-interpreting tapes at the end of the day) reduced the need to call patients back from 30% (27/90) to 2% (2/90). This is likely an overestimate of the benefit of interpreting remotely at the time the study is performed because not all patients requiring additional imaging would necessarily have been called back if the study had been interpreted from a batch at the end of the day. However, the information obtained from these additional images likely decreased the potential for ambiguity in the sonography reports.

Although no significant difference was seen in the time required to perform remote abdominal sonography, the time required to perform remote pelvic sonography was significantly longer than that for pelvic sonography performed for on-site interpretation. This finding was true even though additional images were more often requested by the on-site radiologist (81%) than by the remote radiologist (57%). One explanation for this is that we routinely perform transvaginal imaging in essentially all female patients. If a remote radiologist requests additional images of the pelvis, the patient must be repositioned, the transducer probe reinserted, and imaging performed. If the on-site radiologist obtains additional images, the imaging is done at the end of the initial examination.

Another important factor contributing to the time issue is that our system is designed so that the remote radiologist has to retrieve images from a laser printer after each study and after each additional image requested and then reestablish telephone contact with the on-site technologist after each step, in addition to his or her other clinical duties at the academic medical center. Thus, if the radiologist is called to see another patient at the academic medical center while the images are being transferred, there may be a lag period before the radiologist can contact the technologist at the community-based imaging center and review the case. This time is included in the calculated time per case. The radiologist's busy schedule at the academic medical center may preclude a complete review of all cases before the remote radiologist allows the patient to leave the community-based imaging center. This lack of complete review may explain why two questionable lesions (renal mass and hypoechoic focus in the testes) were not seen before the patient left the community-based imaging center and why both these patients had to return for a second examination.

One limitation of our study is the lack of follow-up in 69 (43%) of 161 patients. However, only 15 (22%) of the 69 patients had abnormal findings on sonography. These figures suggest that a major reason for lack of follow-up was that no follow-up was thought to be indicated by the physician ordering the study because of the normal findings. One may argue that, because sonography is operator-dependent, if the technologist "missed" a lesion, no images of that lesion would have been obtained, and the study would have been interpreted by the radiologist as having normal findings. To reduce this risk at our institution, we always scan patients according to strict protocol, and all images are reviewed by the radiologist for completeness. This practice ensures that all relevant organs are correctly and completely imaged. However, because follow-up was not obtained in all patients, we do not know exactly how many false-negative studies may have actually occurred in our study.

The literature indicates that the rate by which technologists miss lesions varies. In the study of Chan et al. [3], a second sonographic examination performed by a radiologist showed a new abnormality in 183 (12%) of 1510 cases. However, Tessler et al. [4] reported that second sonograms obtained by a radiologist detected a new abnormality in only 28 (7%) of 396 of patients. In a study of pelvic sonography in which on-site and remote interpretations of pelvic sonograms were compared, Rosen et al. [1] reported that the radiologist visualized an abnormality that was not detected by the sonographer in two (2.5%) of 80 cases. Thus, although it is possible that our sonographer failed to visualize an abnormality, we think that is unlikely. In addition, we believe that the rate at which sonographers fail to visualize an abnormality is not well established in the literature. Strict adherence to protocol can help prevent non-visualization of abnormalities.

A second limitation of our study is that the on-site and remote radiologists had different levels of training in sonography. All the remote radiologists had fellowship training in sonography, whereas the on-site radiologists did not. However, we do not think this affected our results. The purpose of our study was to evaluate the performance of remote sonographic interpretation in a "real world" setting. A "cleaner" study design would have involved having the same group of radiologists (general radiologists or sonologists) perform the examination at the community-based imaging center and then, after a period of time so that they were unlikely to recall the case, have the same radiologists reinterpret the static images. This study design would have more accurately reflected differences in sensitivity and specificity directly attributable to on-site versus remote interpretation than a study that introduces the potential bias of observer experience. However, such a study design would not have captured differences in the time required to perform on-site versus remote interpretations.

In conclusion, we believe that remote sonographic interpretation using a laser printer can be done with high levels of diagnostic accuracy in actual clinical practice. Abdominal and pelvic sonograms, including those targeted to the endometrium, can also be safely interpreted remotely. Our study supports one method of providing safe and accurate imaging services to a community-based center.


References
Top
Abstract
Introduction
Subjects and Methods
Results
Discussion
References
 

  1. Rosen MP, Levine D, Carpenter J, et al. Diagnostic accuracy of remote vs. on-site ultrasound interpretation. Radiology 1999;210:733 -736[Abstract/Free Full Text]
  2. Zhou X. Correcting for verification bias in studies of a diagnostic test's accuracy. Stat Methods Med Res 1998;7:337 -353[Abstract/Free Full Text]
  3. Chan V, Hanbidge A, Wilson S, Pron G, Moore L. Case for active physician involvement in US practice. Radiology 1996;199:555 -560[Abstract/Free Full Text]
  4. Tessler FN, Tublin ME, Peters JC, Jie T, Peters TL. Value of selective second-look sonography by radiologists. Radiology 1996;199:551 -553[Abstract/Free Full Text]
  5. Fisk NM, Drysdale K, Garner P, Kyle P, Carvalho JS. Fetal telemedicine: six month pilot of real-time ultrasound and video consultation between the Isle of Wight and London. Br J Obstet Gynaecol 1996;103:1092 -1095[Medline]
  6. Beard DV, Hemminger BM, Keefe B, Mittelstaedt C, Pisano ED, Lee JK. Real-time radiologist review of remote ultrasound using low-cost video and voice. Invest Radiol 1993;28:732 -734[Medline]

Add to CiteULike CiteULike   Add to Complore Complore   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us   Add to Digg Digg   Add to Reddit Reddit   Add to Technorati Technorati    What's this?



This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Rosen, M. P.
Right arrow Articles by Levine, D.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Rosen, M. P.
Right arrow Articles by Levine, D.
Social Bookmarking
 Add to CiteULike   Add to Complore   Add to Connotea   Add to Del.icio.us   Add to Digg   Add to Reddit   Add to Technorati  
What's this?


HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS