Evaluation of an Interactive Science Publishing Tool: Toward Enabling Three-Dimensional Analysis of Medical Images Daniel Rinewalt, MD, Betsy W. Williams, PhD, MPH, Anthony P. Reeves, PhD, Palmi Shah, MD, Edward Hong, MD, James L. Mulshine, MD Rationale and Objectives: Higher resolution medical imaging platforms are rapidly emerging, but there is a challenge in applying these tools in a clinically meaningful way. The purpose of the current study was to evaluate a novel three-dimensional (3D) software imaging environment, known as interactive science publishing (ISP), in appraising 3D computed tomography images and to compare this approach with traditional planar (2D) imaging in a series of lung cancer cases. Materials and Methods: Twenty-four physician volunteers at different levels of training across multiple specialties were recruited to evaluate eight lung cancer–related clinical vignettes. The volunteers were asked to compare the performance of traditional 2D versus the ISP 3D imaging in assessing different visualization environments for diagnostic and measurement processes and to further evaluate the ISP tool in terms of general satisfaction, usability, and probable applicability. Results: Volunteers were satisfied with both imaging methods; however, the 3D environment had significantly higher ratings. Measurement performance was comparable using both traditional 2D and 3D image evaluation. Physicians not trained in 2D measurement approaches versus those with such training demonstrated better performance with ISP and preferred working in the ISP environment. Conclusions: Recent postgraduates with only modest self-administered training performed equally well on 3D and 2D cases. This suggests that the 3D environment has no reduction in accuracy over the conventional 2D approach, while providing the advantage of a digital environment for cross-disciplinary interaction for shared problem solving. Exploration of more effective, efficient, selfdirected training could potentially result in further improvement in image evaluation proficiency and potentially decrease training costs. Key Words: Self-directed learning; 3D imaging; lung cancer diagnosis; health care education. ªAUR, 2015
O
ver the past decade, medical imaging has undergone exponential improvements; however, the ability to use the detailed, textured outputs of these tools requires an understanding of how to manipulate digital imaging data to ensure correct interpretation in guiding clinical management (1–3). This area of image processing is thought to be the province of radiologists, but ongoing trends suggest a greater inclusion of many medical disciplines in applying highresolution imaging in clinical management. Yet the foundational
Acad Radiol 2015; 22:380–386 From the Department of General Surgery, Rush University Medical Center, 1735 West Harrison Street, Suite 206, Chicago, IL 60612 (D.R.); Departments of Behavioral Sciences (B.W.W.) and Psychiatry (B.W.W.), Office of Continuing Medical Education, Rush University Medical Center, Chicago, Illinois; School of Electrical and Computer Engineering, Cornell University, Ithaca, New York (A.P.R.); Department of Radiology, Rush University Medical Center, Chicago, Illinois (P.S.); Department of Cardiovascular-Thoracic Surgery, Rush University Medical Center, Chicago, Illinois (E.H.); and Division of Hematology Oncology, Department of Internal Medicine, Rush University Medical Center, Chicago, Illinois (J.L.M.). Received June 23, 2014; accepted September 23, 2014. Address correspondence to: D.R. e-mail: [email protected] ªAUR, 2015 http://dx.doi.org/10.1016/j.acra.2014.09.012
380
knowledge of proper application of quantitative imaging tools in medical curriculum has not yet emerged as a major focus. Although the technological evolution of medical imaging is moving rapidly, the integration of this new and vast source of digital data into meaningful clinical application is slow. This integration mismatch relates to a number of issues, including pressure to accelerate standard workflow and the inertial influence of shrinking reimbursement. However, the greatest current barrier to implementing image processing approaches may be that clinicians have not been educated about the potential of quantitative imaging. Thus, a knowledge gap exists in how best to apply robust imaging techniques to change the practice of medicine. The interactive science publishing (ISP) tool was released in 2008 as a companion application that is launched by PDF reader software, making it free and easily accessible (4). ISP was designed with functionality for a wide range of imaging file formats to allow for interactive rendering, reformatting, annotation, three-dimensional (3D) measurement, animation, storage, and capture/print of volumetric and image data. The program was formally introduced in the Optics Express Journal in October 2008, with seven articles containing interactive data sets for 3D viewing
Academic Radiology, Vol 22, No 3, March 2015
along with links to obtain the ISP tool (4). Given the potential of ISP to more fully convey imaging concepts and facilitate research, it is important to assess the system to see how the potential user community perceives the value of this resource. We designed and conducted a survey of clinician attitudes at Rush University to determine how they perceive the value of conventional versus new quantitative imaging tools. Lung cancer imaging case studies were used in the survey because imaging in lung cancer is such a dynamic area of interest (1). The survey questions assessed the level of performance in determining diagnoses and volumetric measurements of lung masses using digital imaging and communications in medicine (DICOM) files of computed tomography (CT) images presented in a conventional 2D PDF format versus a 3D environment. To enable the comparison between 2D and 3D performance, we used the ISP imaging environment, which includes both 2D and 3D visualization tools and allows clinicians to digitally manipulate and quantitatively assess medical images. The main goals of the survey were to 1) understand the prior experience of the clinicians; 2) determine clinician satisfaction with each of the data and image presentation methods; and 3) gage the general receptiveness of clinicians to use and master a representative panel of image processing tools for lung cancer clinical management. Essentially, we describe a visualization environment designed to support the use of image analysis tools to assist in the quantitative assessment of 3D medical images. We then evaluated this environment to determine user assessment of whether these tools to better understand physicians’ acceptance of these tools in allowing greater integration of image analysis into clinical management.
MATERIALS AND METHODS We enrolled 24 uncompensated volunteers to participate in a pilot survey comparing ISP software with the more traditional 2D approach. The study was exempt from formal institutional review board process as no identifiable patient information was recorded, and all participants were physicians at Rush University. The study consisted of a survey with two primary components. The first component was a qualitative or quantitative investigation, with volunteers using the 2D PDF or 2D/3D ISP software to make judgments about eight clinical vignettes. The second component captured the participants’ perception about their experience using the ISP software. Training tools were prepared as electronic instructions for the two imaging environments (2D PDF or 2D/3D ISP file structure). One research coordinator administered all evaluations.
EVALUATION OF THE ISP TOOL
The eight vignettes were divided into two sets of four, with the first set focused on diagnosis (benign or malignant), and the second set focused on overall measurements of change in the size of nodules over time. Each set (four vignettes) was completed with two cases using traditional 2D PDF format and the remainder using the new interactive 2D/3D ISP software. The vignette evaluation was presented in a multiple-choice format. Each question had one correct response of five possible choices. The correct response was determined before the initiation of the survey process by two of the authors (J.L.M. and A.P.R.). Results of the analysis, which were generated by coding each correct response as 2 and each incorrect response as 1, were displayed as bar graphs representing the average performance in each condition. Process Evaluation
Volunteers were also asked to evaluate the ISP tool by rating a series of statements concerning general satisfaction, usability, and probable application of the tool. All answers were provided using a Likert scale with ‘‘strongly disagree’’ rated as 1 and ‘‘strongly agree’’ rated as 5. Statistical Analysis
The core analysis undertaken was an analysis of variance to determine if the imaging method (ISP vs. traditional) had a significant effect on the participants’ performance in either diagnosis or measurement of lesions. A secondary mean difference test (t test) was undertaken to investigate the participants’ relative satisfaction with the two imaging approaches. As a result of the findings, post hoc descriptive analyses were undertaken. These analyses used a factor analysis of preference data to abstract the core characteristics of the participants’ preference for either traditional or ISP imaging approaches (two factors were retained). These data were used in a cluster analysis to identify the types of participants by preference for imaging method. The characteristics of these clusters were then described to help elucidate the characteristics of physicians as they relate to imaging preference.
RESULTS The 24 participants represented a wide variety of specialties, including internal medicine (59%), general surgery (13%), anesthesia (9%), and the rest all at 4% including cardiology, neurology, and cardiothoracic surgery (Fig 1). The participants ranged from residents to senior faculty/practitioners. Most participants were at an early stage in their career.
Quantitative Evaluation Vignettes
Quantitative Evaluation of Clinical Vignettes
The vignette evaluation included eight clinical cases on the basis of incidentally discovered lung nodules observed using spiral CT in patients at high risk of developing lung cancer.
Four vignettes involved analysis of diagnostic features, and the remaining four involved measurement of volume change in response to drug therapy. The primary finding was that, in 381
RINEWALT ET AL
Academic Radiology, Vol 22, No 3, March 2015
multivariate effect of method (F = 401.113; df(3/44); P < .05). Table 1 presents the full set of means. Combined Analysis
Figure 1. Distribution of specialties.
general, performance on the diagnostic questions was essentially identical for ISP and traditional imagery, whereas performance on the measurement questions was higher using the traditional PDF format. The statistical analysis used a repeated-measures analysis of variance approach (ie, 2 [question focus] 2 [imaging method] 2 [question]). All the main effects and all the interactions were statistically significant (P < .05). The results of the analysis, which were generated by coding each correct response as 2 and each incorrect response as 1, are shown in Figure 2; the bars represent the average performance in each condition. The Question Focus conditions are indicated as ‘‘Diagnosis’’ (for questions related to a diagnosis vignette) and as ‘‘Measurement’’ (for questions related a therapeutic response vignette). As shown, performance was better on diagnosis-related questions than measurement-related questions. The designations 2D versus ISP in Figure 2 indicate questions within a Question Focus that allow the respondent to use the traditional PDF format versus ISP, respectively. As can be seen, in the diagnosis condition, there is essentially no difference in participant performance between the two imaging systems; however, in the measurement condition, the participants performed significantly better using the traditional PDF format than ISP. The grand mean of responses was approximately 3.5, potentially indicating a slight positive bias on the part of the raters. The general statement, ‘‘I am satisfied with ISP information system as a source of information, data, and analysis for my work or research,’’ had an overall mean of approximately 3.71, which was higher than the grand mean. The corresponding statement, ‘‘I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my clinical practice,’’ had an overall mean of 3.46. The two findings were not statistically significantly different from one another (P = .263). In general, the participants were satisfied with both communication methods; however, across satisfaction measures, ISP had higher ratings than did the traditional presentation. When combined and analyzed, the preference for ISP as a presentation method was confirmed by an overall significant 382
In an effort to understand these rather counterintuitive results (ie, that participants performed significantly worse using the tool for which they had a significantly higher preference), an additional review of preference and performance data was undertaken in a combined analysis. The findings should be viewed as very preliminary as the number of respondents was limited for this type of descriptive analysis. The second review focused on measurement-related questions, as the first set of data showed no apparent differences between the two imaging methods in performance on the diagnostic questions in the clinical diagnosis vignettes. The participants had differing levels of experience, ranging from the first-year residents to mature practitioners. As shown in Figure 3, participants were divided into three groups on the basis of length of training: group 1 had 1 year or less of postgraduate training, group 2 had less than 5 years of postgraduate training, and group 3 had 5 years or more of postgraduate training. An analysis of the relationship between the extent of experience and performance showed that as experience increases, performance improves. Given that differences exist in the frequency of utilization of imaging data by specialty, an analysis was undertaken that divided the participants into two groups—diagnostic specialty versus procedural specialty. No consistent pattern of finding was discernible from this analysis. The relationship between attitudes toward imaging methods and performance on the vignettes was also analyzed. The data on satisfaction with each of the two methodologies were consolidated into two main factors. One factor measured preference for the ISP methodology, and the other factor measured antipathy toward the traditional methodology. By using these factors, three types of respondents were developed and described as follows: ISP Philic (Like ISP-Neutral 2-D); ISP Phobic (Mild Dislike ISP-Neutral 2-D); and 2-D Phobic (Like ISP-Dislike 2-D). Figure 4 presents a cluster map of the three clusters, with factor score scales provide for the placement; of note, some caution should be applied in the interpretation. The dislike (ISP Phobic) scores for ISP are less than the difference appears, as factor scores are based on normalized data, and the actual mean of this distribution is high; therefore, the ‘‘phobic’’ description is more correctly ‘‘less philic.’’ Similarly, 2-D Phobic scores (although correct) represent an understatement in that the range is skewed low, with positive scores relatively mild compared to ISP ratings. These clusters were then used to review performance data. To create a single informative measure, a difference score was developed for each participant by subtracting their 2D performance scores from their ISP performance scores. These results were then transformed so that the scores were positive. Thus, on this composite measure of performance, the higher the relative performance on ISP, the higher the score. As shown
Academic Radiology, Vol 22, No 3, March 2015
EVALUATION OF THE ISP TOOL
Figure 2. Summary of survey responses for clinical vignettes.
TABLE 1. Applicant Questionnaire Question 1. The Interactive Science Publishing (ISP) represents a major step forward in providing a setting to interact with visual clinical information. 2. I am satisfied with ISP information system as a source of information, data, and analysis for my work or research. 3. I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my clinical practice. 4. I am satisfied with ISP information system as a source of information, data, and analysis for my research reading. 5. I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my research reading. 6. I am satisfied with the visualization tools for interacting with the clinical images provided by the ISP information system. 7. I am satisfied with the visualization tools for interacting with clinical images provided by traditional electronic journals. 8. If available I would make use of the ISP technology in preference to the traditional technology. 9. I understand the application of the image processing tools available in the ISP to quantify aspects of the clinical image. 10. I am satisfied with the computer interface provided to support the ISP. 11. The directions and training on the interactive Science Publishing (ISP) are easy to understand. 12. The interactive Science Publishing instrument is easy to use. 13. Using the ISP instrument I am more confident of my clinical judgment. 14. The ISP will prove useful in my patient care or research activities. 15. The ISP will prove useful in my research projects.
in Figure 5, the dominant determinant of relative performance on ISP was the degree to which the participant was dissatisfied with the current 2D imaging technology. An analysis to determine the relationship between training and cluster membership was also undertaken. Of interest, no member in the group with the longest training fell into the 2-D Phobic cluster. The training group with the largest percentage of 2-D Phobic members was the one in which members were in the middle of or just finishing their postgraduate training. The groups with the largest percentage of ISP Philic
Mean
Standard Deviation
3.67
0.8165
3.71
0.7506
3.46
0.7790
3.79
0.7211
3.58
0.8805
3.63
1.0135
3.46
0.8836
3.67 3.42
0.8681 1.0598
3.79 3.75 3.79 3.17 3.46 3.25
0.8836 1.0734 1.0206 0.8165 0.9771 0.7940
members were those composed of beginning residents and residents late in the training process or nearing completion of their residency.
DISCUSSION Radiologists generally approach the analysis of a clinical image with the intention of performing a qualitative analysis. Examples of this 2D analysis include diagnosing a fractured bone, 383
RINEWALT ET AL
Figure 3. Impact of training length on performance. Group 1: 0-1 years of postgraduate training. Group 2: 1-5 years of postgraduate training. Group 3: greater than 5 years of postgraduate training. SCOREB1A2D and SCOREB2A2D: Score on measurement vignettes using 2D. SCOREB1BISP and SCOREB2BISP: Score on measurement vignettes using ISP.
Figure 4. Scatter diagram of preference for visualization environment. ISP, interactive science publishing.
detecting the presence of free air under the diaphragm, or correctly placing the endotrachial tube in the midline. More recently, the exquisite resolution of medical imaging systems allows for much more comprehensive acquisition of imaged structures enabling the quantitation of clinically interesting structures, such as the volume of lung nodules or the extent of occlusion with a vascular structure. The transition from qualitative image assessment to quantitative imaging represents a change for clinical medicine, with profound implications for medical practice, which have not yet attracted general scrutiny. Simply put, the challenge is supporting the analytical framework in moving medical imaging from picture taking to a new discipline of measurement science. The potential for measuring elements within clinical images exists across all medical professions, from caring for patients with atherosclerotic vascular structure to assessing early cognitive decline to assessing drug therapy response. 384
Academic Radiology, Vol 22, No 3, March 2015
Figure 5. Graph of the relationship between performance and preference for imaging methodology. Higher scores on the vertical axis indicate a greater relative score on ISP (interactive science publishing).
In this pilot analysis, the ISP tool was generally preferred over the traditional imaging methodology when analyzed for all measures and across the various ranges of clinical experience and specialty. This is feedback from a cohort of clinicians who may be representative of the overall field of medicine. Although we had hypothesized that clinicians would prefer the newer ISP technology, it is understandable that most volunteer clinicians would favor the traditional imaging methodology on which they had been trained and had the most familiarity. These clinician subjects did perform better on interpretation with the traditional methodology when measuring differences in the size of lung nodules. The fact that this finding correlates in a statistically significantly fashion with experience indicates that medical training is clearly effective at teaching clinicians to use technology that is presently the standard of care. Yet performance of the ISP tool was found to be equal to that of the traditional 2D methodology in the diagnostic vignettes. It should be noted this was true after only a very brief self-taught instructional period using the experimental technique as opposed to multiple years of structured curriculum with 2D technology formulated by experts in medical education. Although the ISP tool did not perform as well for measurement-based tasks overall, this effect was minimal in participants who reported being less comfortable with 2D than 3D imaging. Participants who reported such discomfort were noted to be at an earlier stage of training and tended to have better performance using the ISP than 2D tool. Considered against the backdrop of a teaching system that presents the traditional imaging approach as an embedded aspect of the medical school experience, this transition constitutes a complex process. This analysis was performed on DICOM imaging data from helical CT scans of lung cancer cases. Imaging resolution is
Academic Radiology, Vol 22, No 3, March 2015
particularly favorable with CT in lung tissue, with the soft tissue density of early lung cancers silhouetted by the air density of normal lung tissue. This favorable contrast has allowed CT to improve early lung cancer detection, and this approach has been recently been demonstrated to result in a significant reduction in lung cancer deaths (4,5). Recently, 3D image processing has been proposed as a tool in the screening evaluation process to confirm the clinical nature of a suspected cancer to determine if the potential tumor is growing (6). This evaluation is performed by obtaining 3D CT images of the tumor across a time interval (usually 3 months). If the tumor has grown by a significant amount (usually >10%), the case is sent for full invasive diagnostic evaluation, as tumor growth is a central hallmark of cancer. This 3D application is an early medical application of quantitative imaging and representative of a larger movement to improve on the current limitations of planar imaging techniques. Although the one-dimensional Response Evaluation Criteria in Solid Tumors (RECIST) approach has gained considerable traction because of its speed and simplicity, it does not fully evaluate the complex geometry of typical cancer lesions (7,8). RECIST assumes that lung nodules routinely change size in a symmetric fashion. It is under these conditions that planar measurement would faithfully reflect the true extent of volume changes by a growing tumor nodule. Unfortunately, studies have revealed that tumors frequently do not grow symmetrically and that substantial variability exists in RECIST measurements (8). Increased precision is necessary as some studies have discovered that RECIST failed to accurately identify up to 33% of patients adequately responding to tyrosine kinase inhibitors (9). By not providing the correct assessment, clinicians may compromise the care of cancer patients. For example, in one case of an asymmetric nodule followed over time with serial CT scans, traditional measures classified the doubling time as benign, whereas 3D-based measurements revealed the doubling time to be consistent with malignancy (7). Concern with such inaccuracies has encouraged movement toward 3D imaging. The development of software like the ISP tool may facilitate the way clinicians integrate quantitative imaging into routine clinical management (10,11). The best example of an analogous situation involving the transition to 3D visualization tools exists in the architectural community where it took several decades to learn how to integrate the enormous amount of digital data generated using computer-aided building design software. The use of 3D digital tool transformed the field of architecture and demonstrates the potential contribution of managing digital data as a resource for shared planning and education locally or across the web. Interestingly, the original motivation for moving architectural planning from traditional blueprints to computer-aided design (CAD) formats was to improve the accuracy of the design documents. However, it was quickly appreciated that CAD planning facilitated collaborative interaction among participants in the design process in profound and unanticipated ways. The notable achievements of Frank Gehry illustrate how CAD emerged as a transformative tool in architecture. The Gehry
EVALUATION OF THE ISP TOOL
design team, which was a late adapter of 3D design, was initially characterized as CAD phobic (12); however, after many years of aversion, they rapidly assimilated process knowledge and soon emerged as radical innovators, producing a series of original (yet already iconic) structures. Key to their success was an understanding that the 3D design process was disruptive but in a different way for different stakeholders. The ability to visualize the entire integrated planned structured in scale allowed simultaneous tailored evolution of numerous construction processes, such as structural engineering, sound management, fire safety, and dry wall fabrication. This digital planning environment resulted in a new environment (or matrix) of innovation by allowing simultaneous interaction with the same digital data by separate specialists involved in the construction planning process. Each construction discipline solved its respective planning problem by interacting with the shared digital data that were constrained by the dynamic changes to the building plans added in real time by all the disciplines of building designers. This type of digital environment provided a level of integration, analysis, and coordination that had not been previously feasible. Furthermore, architecture is not the only field in which 3D planning has been transformational. Another example of this disruptive potential exists with the integration of 3D industrial fabrication into the progress in computer chip design which made possible a stacked rather than planar chip designs allowing significant improvement with heat dissipation (13). The lesson of the architectural experience with 3D innovation is that this type of digital planning and analysis capability is most relevant to multidisciplinary medical problem solving. Typically, in volumetric imaging applications, radiologists are focused on accurate and robust measurement of imaging findings to address a diagnostic question. A surgeon uses 3D imaging information functionally to guide therapeutic interventions. A neurosurgeon, for example, uses 3D imaging to assess luminal patency to assist in the selection of the proper interventional device for effective evacuation of the occluding clot from a vascular structure to limit the extent of damage from an evolving stroke. Similarly, 3D imaging has enabled a new functional classification of spinal scoliosis on the basis of the nature of torsional stress across spinous structures—a perspective that was not possible with the use of 2D spinal images (14). The pace of biomedical evolution may accelerate with the use of 3D data sets as communication/collaboration tools. This was the goal of the Optical Society in developing ISP to serve as such a reference visualization resource to facilitate research collaboration (4). ISP can serve as an infrastructure to systematically acquire and host libraries of medical images as a Web-accessible resource. In certain instances, libraries of image data sets may also include linked clinical outcome data from the individuals who donated the medical images. Such image/clinical outcome libraries constitute an invaluable resource for image software development and other types of research and education that could be scaled using hosted Web architecture ensuring that this resource would be economically accessible to the global medical community. 385
RINEWALT ET AL
Beyond clinical applications of CT, the underlying 3D rendering and related visualization approaches are being applied to basic studies of subcellular structures (15). The ISP environment may serve as a new tool to enable collaboration to scale the divide between basic and clinical science research. This ISP collaboration tool, which enables shared access to primary imaging data across all disciplines while providing a toolkit of reference for quantitative imaging techniques, may represent a new foundational resource for research convergence (16). From this pilot evaluation, ISP software application seems to be a useful tool for viewing and interpreting radiographic images and provides a strategic resource for serving as a collaboration tool to allow many groups to interact dynamically with the same digital (DICOM) imaging data set. As the ISP tool was found to have minimal (if any) diminution in accuracy compared to traditional methods and was favored by new users, we propose that increased use of this technology is warranted. Further analysis of user perspectives around the 3D innovation process may assist with catalyzing progress with imaging while also providing an opportunity to better understand and facilitate the process of innovation. ACKNOWLEDGMENT The authors wish to thank Ricardo Avila for his critical assistance and technical advice in the development of the interactive science publishing evaluation tool and Josephine Volgi, RN, for her gracious perseverance in administering all the evaluations. The authors thank Artit Jirapatnakul for developing the PDF vehicle to administer the survey tool in the defined sequence while allowing recovery of the evaluation responses. They also thank David Shersher for his help with identifying volunteers for this effort and the many Rush physicians, who volunteered their time to participate in the evaluation process. The authors appreciate the support for the research and development of this article by a contract with the Optical Society of America and the
386
Academic Radiology, Vol 22, No 3, March 2015
US Air Force. We appreciate the critical review of the article by Dr. Thomas Baer, Photonic Research Center at Stanford University.
REFERENCES 1. Mulshine JL, Avila R, Yankelevitz D, et al. Use of high resolution CT imaging data in lung cancer drug development: measuring progress. Oncology 2009; 23:434–438. 2. Buckler AJ, Mulshine JL, Gottlieb R, et al. The use of volumetric CT as an imaging biomarker in lung cancer. Acad Radiol 2010; 17:100–106. 3. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009; 45:228–247. 4. Mulshine JL, Baer TM, Avila RS. Introduction: imaging in diagnosis and treatment of lung cancer. Opt Express 2010; 18:15242–15243. 5. Aberle DR, Adams AM, Berg CD, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011; 365: 395–409. 6. Mulshine JL, Jablons DM. Volume CT for diagnostics work-up in a randomized lung cancer screening trial: learning from Nelson. N Engl J Med 2009; 361:2221–2229. 7. Yankelevitz DF, Reeves AP, Kostis WJ, et al. Small pulmonary nodules: volumetrically determined growth rates based on CT evaluation. Radiology 2000; 217:251–256. 8. Gavrielides MA, Kinnard LM, Myers KJ, et al. Noncalcified lung nodules: volumetric assessment with thoracic CT. Radiology 2009; 251:26–37. 9. Zhao B, Oxnard GR, Moskowitz CS, et al. A pilot study of volume measurement as a method of tumor response evaluation to aid biomarker development. Clin Cancer Res 2010; 16:4647–4653. 10. Revel M, Lefort C, Bissery A, et al. Pulmonary nodules: preliminary experience with three-dimensional evaluation. Radiology 2004; 231:459–466. 11. Krishnan K, Ibanez L, Turner WD, et al. An open-source toolkit for the volumetric measurement of CT lung lesions. Optics Express 2010; 18: 15256–15266. 12. Boland RJ, Lyytinen K, Yoo Y. Wakes of innovation in project networks: the case of digital 3-D representations in architecture, engineering, and construction. Organization Science 2007; 18:631–647. 13. Young AM, Koester SJ. 3D process technology considerations. In: Xie Y, Cong J, Sapatnekar S, eds. Three-dimensional integrated circuit design, integrated circuits and systems. New York, NY: Springer LLC, 2010. 14. Labelle H, Aubin CE, Jackson R, et al. Seeing the spine in 3D: how will it change what we do? J Pediatr Orthop 2011; 31:S37–S45. 15. Miao Q, Yu J, Rahn JR, et al. Dual-mode optical projection tomography microscope using gold nanorods and hematoxylin-stained cancer cells. Opt Lett 2010; 35:1037–1039. 16. Sharp PA, Langer R. Promoting convergence in biomedical science. Science 2011; 333:527.
O
ver the past decade, medical imaging has undergone exponential improvements; however, the ability to use the detailed, textured outputs of these tools requires an understanding of how to manipulate digital imaging data to ensure correct interpretation in guiding clinical management (1–3). This area of image processing is thought to be the province of radiologists, but ongoing trends suggest a greater inclusion of many medical disciplines in applying highresolution imaging in clinical management. Yet the foundational
Acad Radiol 2015; 22:380–386 From the Department of General Surgery, Rush University Medical Center, 1735 West Harrison Street, Suite 206, Chicago, IL 60612 (D.R.); Departments of Behavioral Sciences (B.W.W.) and Psychiatry (B.W.W.), Office of Continuing Medical Education, Rush University Medical Center, Chicago, Illinois; School of Electrical and Computer Engineering, Cornell University, Ithaca, New York (A.P.R.); Department of Radiology, Rush University Medical Center, Chicago, Illinois (P.S.); Department of Cardiovascular-Thoracic Surgery, Rush University Medical Center, Chicago, Illinois (E.H.); and Division of Hematology Oncology, Department of Internal Medicine, Rush University Medical Center, Chicago, Illinois (J.L.M.). Received June 23, 2014; accepted September 23, 2014. Address correspondence to: D.R. e-mail: [email protected] ªAUR, 2015 http://dx.doi.org/10.1016/j.acra.2014.09.012
380
knowledge of proper application of quantitative imaging tools in medical curriculum has not yet emerged as a major focus. Although the technological evolution of medical imaging is moving rapidly, the integration of this new and vast source of digital data into meaningful clinical application is slow. This integration mismatch relates to a number of issues, including pressure to accelerate standard workflow and the inertial influence of shrinking reimbursement. However, the greatest current barrier to implementing image processing approaches may be that clinicians have not been educated about the potential of quantitative imaging. Thus, a knowledge gap exists in how best to apply robust imaging techniques to change the practice of medicine. The interactive science publishing (ISP) tool was released in 2008 as a companion application that is launched by PDF reader software, making it free and easily accessible (4). ISP was designed with functionality for a wide range of imaging file formats to allow for interactive rendering, reformatting, annotation, three-dimensional (3D) measurement, animation, storage, and capture/print of volumetric and image data. The program was formally introduced in the Optics Express Journal in October 2008, with seven articles containing interactive data sets for 3D viewing
Academic Radiology, Vol 22, No 3, March 2015
along with links to obtain the ISP tool (4). Given the potential of ISP to more fully convey imaging concepts and facilitate research, it is important to assess the system to see how the potential user community perceives the value of this resource. We designed and conducted a survey of clinician attitudes at Rush University to determine how they perceive the value of conventional versus new quantitative imaging tools. Lung cancer imaging case studies were used in the survey because imaging in lung cancer is such a dynamic area of interest (1). The survey questions assessed the level of performance in determining diagnoses and volumetric measurements of lung masses using digital imaging and communications in medicine (DICOM) files of computed tomography (CT) images presented in a conventional 2D PDF format versus a 3D environment. To enable the comparison between 2D and 3D performance, we used the ISP imaging environment, which includes both 2D and 3D visualization tools and allows clinicians to digitally manipulate and quantitatively assess medical images. The main goals of the survey were to 1) understand the prior experience of the clinicians; 2) determine clinician satisfaction with each of the data and image presentation methods; and 3) gage the general receptiveness of clinicians to use and master a representative panel of image processing tools for lung cancer clinical management. Essentially, we describe a visualization environment designed to support the use of image analysis tools to assist in the quantitative assessment of 3D medical images. We then evaluated this environment to determine user assessment of whether these tools to better understand physicians’ acceptance of these tools in allowing greater integration of image analysis into clinical management.
MATERIALS AND METHODS We enrolled 24 uncompensated volunteers to participate in a pilot survey comparing ISP software with the more traditional 2D approach. The study was exempt from formal institutional review board process as no identifiable patient information was recorded, and all participants were physicians at Rush University. The study consisted of a survey with two primary components. The first component was a qualitative or quantitative investigation, with volunteers using the 2D PDF or 2D/3D ISP software to make judgments about eight clinical vignettes. The second component captured the participants’ perception about their experience using the ISP software. Training tools were prepared as electronic instructions for the two imaging environments (2D PDF or 2D/3D ISP file structure). One research coordinator administered all evaluations.
EVALUATION OF THE ISP TOOL
The eight vignettes were divided into two sets of four, with the first set focused on diagnosis (benign or malignant), and the second set focused on overall measurements of change in the size of nodules over time. Each set (four vignettes) was completed with two cases using traditional 2D PDF format and the remainder using the new interactive 2D/3D ISP software. The vignette evaluation was presented in a multiple-choice format. Each question had one correct response of five possible choices. The correct response was determined before the initiation of the survey process by two of the authors (J.L.M. and A.P.R.). Results of the analysis, which were generated by coding each correct response as 2 and each incorrect response as 1, were displayed as bar graphs representing the average performance in each condition. Process Evaluation
Volunteers were also asked to evaluate the ISP tool by rating a series of statements concerning general satisfaction, usability, and probable application of the tool. All answers were provided using a Likert scale with ‘‘strongly disagree’’ rated as 1 and ‘‘strongly agree’’ rated as 5. Statistical Analysis
The core analysis undertaken was an analysis of variance to determine if the imaging method (ISP vs. traditional) had a significant effect on the participants’ performance in either diagnosis or measurement of lesions. A secondary mean difference test (t test) was undertaken to investigate the participants’ relative satisfaction with the two imaging approaches. As a result of the findings, post hoc descriptive analyses were undertaken. These analyses used a factor analysis of preference data to abstract the core characteristics of the participants’ preference for either traditional or ISP imaging approaches (two factors were retained). These data were used in a cluster analysis to identify the types of participants by preference for imaging method. The characteristics of these clusters were then described to help elucidate the characteristics of physicians as they relate to imaging preference.
RESULTS The 24 participants represented a wide variety of specialties, including internal medicine (59%), general surgery (13%), anesthesia (9%), and the rest all at 4% including cardiology, neurology, and cardiothoracic surgery (Fig 1). The participants ranged from residents to senior faculty/practitioners. Most participants were at an early stage in their career.
Quantitative Evaluation Vignettes
Quantitative Evaluation of Clinical Vignettes
The vignette evaluation included eight clinical cases on the basis of incidentally discovered lung nodules observed using spiral CT in patients at high risk of developing lung cancer.
Four vignettes involved analysis of diagnostic features, and the remaining four involved measurement of volume change in response to drug therapy. The primary finding was that, in 381
RINEWALT ET AL
Academic Radiology, Vol 22, No 3, March 2015
multivariate effect of method (F = 401.113; df(3/44); P < .05). Table 1 presents the full set of means. Combined Analysis
Figure 1. Distribution of specialties.
general, performance on the diagnostic questions was essentially identical for ISP and traditional imagery, whereas performance on the measurement questions was higher using the traditional PDF format. The statistical analysis used a repeated-measures analysis of variance approach (ie, 2 [question focus] 2 [imaging method] 2 [question]). All the main effects and all the interactions were statistically significant (P < .05). The results of the analysis, which were generated by coding each correct response as 2 and each incorrect response as 1, are shown in Figure 2; the bars represent the average performance in each condition. The Question Focus conditions are indicated as ‘‘Diagnosis’’ (for questions related to a diagnosis vignette) and as ‘‘Measurement’’ (for questions related a therapeutic response vignette). As shown, performance was better on diagnosis-related questions than measurement-related questions. The designations 2D versus ISP in Figure 2 indicate questions within a Question Focus that allow the respondent to use the traditional PDF format versus ISP, respectively. As can be seen, in the diagnosis condition, there is essentially no difference in participant performance between the two imaging systems; however, in the measurement condition, the participants performed significantly better using the traditional PDF format than ISP. The grand mean of responses was approximately 3.5, potentially indicating a slight positive bias on the part of the raters. The general statement, ‘‘I am satisfied with ISP information system as a source of information, data, and analysis for my work or research,’’ had an overall mean of approximately 3.71, which was higher than the grand mean. The corresponding statement, ‘‘I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my clinical practice,’’ had an overall mean of 3.46. The two findings were not statistically significantly different from one another (P = .263). In general, the participants were satisfied with both communication methods; however, across satisfaction measures, ISP had higher ratings than did the traditional presentation. When combined and analyzed, the preference for ISP as a presentation method was confirmed by an overall significant 382
In an effort to understand these rather counterintuitive results (ie, that participants performed significantly worse using the tool for which they had a significantly higher preference), an additional review of preference and performance data was undertaken in a combined analysis. The findings should be viewed as very preliminary as the number of respondents was limited for this type of descriptive analysis. The second review focused on measurement-related questions, as the first set of data showed no apparent differences between the two imaging methods in performance on the diagnostic questions in the clinical diagnosis vignettes. The participants had differing levels of experience, ranging from the first-year residents to mature practitioners. As shown in Figure 3, participants were divided into three groups on the basis of length of training: group 1 had 1 year or less of postgraduate training, group 2 had less than 5 years of postgraduate training, and group 3 had 5 years or more of postgraduate training. An analysis of the relationship between the extent of experience and performance showed that as experience increases, performance improves. Given that differences exist in the frequency of utilization of imaging data by specialty, an analysis was undertaken that divided the participants into two groups—diagnostic specialty versus procedural specialty. No consistent pattern of finding was discernible from this analysis. The relationship between attitudes toward imaging methods and performance on the vignettes was also analyzed. The data on satisfaction with each of the two methodologies were consolidated into two main factors. One factor measured preference for the ISP methodology, and the other factor measured antipathy toward the traditional methodology. By using these factors, three types of respondents were developed and described as follows: ISP Philic (Like ISP-Neutral 2-D); ISP Phobic (Mild Dislike ISP-Neutral 2-D); and 2-D Phobic (Like ISP-Dislike 2-D). Figure 4 presents a cluster map of the three clusters, with factor score scales provide for the placement; of note, some caution should be applied in the interpretation. The dislike (ISP Phobic) scores for ISP are less than the difference appears, as factor scores are based on normalized data, and the actual mean of this distribution is high; therefore, the ‘‘phobic’’ description is more correctly ‘‘less philic.’’ Similarly, 2-D Phobic scores (although correct) represent an understatement in that the range is skewed low, with positive scores relatively mild compared to ISP ratings. These clusters were then used to review performance data. To create a single informative measure, a difference score was developed for each participant by subtracting their 2D performance scores from their ISP performance scores. These results were then transformed so that the scores were positive. Thus, on this composite measure of performance, the higher the relative performance on ISP, the higher the score. As shown
Academic Radiology, Vol 22, No 3, March 2015
EVALUATION OF THE ISP TOOL
Figure 2. Summary of survey responses for clinical vignettes.
TABLE 1. Applicant Questionnaire Question 1. The Interactive Science Publishing (ISP) represents a major step forward in providing a setting to interact with visual clinical information. 2. I am satisfied with ISP information system as a source of information, data, and analysis for my work or research. 3. I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my clinical practice. 4. I am satisfied with ISP information system as a source of information, data, and analysis for my research reading. 5. I am satisfied with traditional electronic journal style information as a source of information, data, and analysis for my research reading. 6. I am satisfied with the visualization tools for interacting with the clinical images provided by the ISP information system. 7. I am satisfied with the visualization tools for interacting with clinical images provided by traditional electronic journals. 8. If available I would make use of the ISP technology in preference to the traditional technology. 9. I understand the application of the image processing tools available in the ISP to quantify aspects of the clinical image. 10. I am satisfied with the computer interface provided to support the ISP. 11. The directions and training on the interactive Science Publishing (ISP) are easy to understand. 12. The interactive Science Publishing instrument is easy to use. 13. Using the ISP instrument I am more confident of my clinical judgment. 14. The ISP will prove useful in my patient care or research activities. 15. The ISP will prove useful in my research projects.
in Figure 5, the dominant determinant of relative performance on ISP was the degree to which the participant was dissatisfied with the current 2D imaging technology. An analysis to determine the relationship between training and cluster membership was also undertaken. Of interest, no member in the group with the longest training fell into the 2-D Phobic cluster. The training group with the largest percentage of 2-D Phobic members was the one in which members were in the middle of or just finishing their postgraduate training. The groups with the largest percentage of ISP Philic
Mean
Standard Deviation
3.67
0.8165
3.71
0.7506
3.46
0.7790
3.79
0.7211
3.58
0.8805
3.63
1.0135
3.46
0.8836
3.67 3.42
0.8681 1.0598
3.79 3.75 3.79 3.17 3.46 3.25
0.8836 1.0734 1.0206 0.8165 0.9771 0.7940
members were those composed of beginning residents and residents late in the training process or nearing completion of their residency.
DISCUSSION Radiologists generally approach the analysis of a clinical image with the intention of performing a qualitative analysis. Examples of this 2D analysis include diagnosing a fractured bone, 383
RINEWALT ET AL
Figure 3. Impact of training length on performance. Group 1: 0-1 years of postgraduate training. Group 2: 1-5 years of postgraduate training. Group 3: greater than 5 years of postgraduate training. SCOREB1A2D and SCOREB2A2D: Score on measurement vignettes using 2D. SCOREB1BISP and SCOREB2BISP: Score on measurement vignettes using ISP.
Figure 4. Scatter diagram of preference for visualization environment. ISP, interactive science publishing.
detecting the presence of free air under the diaphragm, or correctly placing the endotrachial tube in the midline. More recently, the exquisite resolution of medical imaging systems allows for much more comprehensive acquisition of imaged structures enabling the quantitation of clinically interesting structures, such as the volume of lung nodules or the extent of occlusion with a vascular structure. The transition from qualitative image assessment to quantitative imaging represents a change for clinical medicine, with profound implications for medical practice, which have not yet attracted general scrutiny. Simply put, the challenge is supporting the analytical framework in moving medical imaging from picture taking to a new discipline of measurement science. The potential for measuring elements within clinical images exists across all medical professions, from caring for patients with atherosclerotic vascular structure to assessing early cognitive decline to assessing drug therapy response. 384
Academic Radiology, Vol 22, No 3, March 2015
Figure 5. Graph of the relationship between performance and preference for imaging methodology. Higher scores on the vertical axis indicate a greater relative score on ISP (interactive science publishing).
In this pilot analysis, the ISP tool was generally preferred over the traditional imaging methodology when analyzed for all measures and across the various ranges of clinical experience and specialty. This is feedback from a cohort of clinicians who may be representative of the overall field of medicine. Although we had hypothesized that clinicians would prefer the newer ISP technology, it is understandable that most volunteer clinicians would favor the traditional imaging methodology on which they had been trained and had the most familiarity. These clinician subjects did perform better on interpretation with the traditional methodology when measuring differences in the size of lung nodules. The fact that this finding correlates in a statistically significantly fashion with experience indicates that medical training is clearly effective at teaching clinicians to use technology that is presently the standard of care. Yet performance of the ISP tool was found to be equal to that of the traditional 2D methodology in the diagnostic vignettes. It should be noted this was true after only a very brief self-taught instructional period using the experimental technique as opposed to multiple years of structured curriculum with 2D technology formulated by experts in medical education. Although the ISP tool did not perform as well for measurement-based tasks overall, this effect was minimal in participants who reported being less comfortable with 2D than 3D imaging. Participants who reported such discomfort were noted to be at an earlier stage of training and tended to have better performance using the ISP than 2D tool. Considered against the backdrop of a teaching system that presents the traditional imaging approach as an embedded aspect of the medical school experience, this transition constitutes a complex process. This analysis was performed on DICOM imaging data from helical CT scans of lung cancer cases. Imaging resolution is
Academic Radiology, Vol 22, No 3, March 2015
particularly favorable with CT in lung tissue, with the soft tissue density of early lung cancers silhouetted by the air density of normal lung tissue. This favorable contrast has allowed CT to improve early lung cancer detection, and this approach has been recently been demonstrated to result in a significant reduction in lung cancer deaths (4,5). Recently, 3D image processing has been proposed as a tool in the screening evaluation process to confirm the clinical nature of a suspected cancer to determine if the potential tumor is growing (6). This evaluation is performed by obtaining 3D CT images of the tumor across a time interval (usually 3 months). If the tumor has grown by a significant amount (usually >10%), the case is sent for full invasive diagnostic evaluation, as tumor growth is a central hallmark of cancer. This 3D application is an early medical application of quantitative imaging and representative of a larger movement to improve on the current limitations of planar imaging techniques. Although the one-dimensional Response Evaluation Criteria in Solid Tumors (RECIST) approach has gained considerable traction because of its speed and simplicity, it does not fully evaluate the complex geometry of typical cancer lesions (7,8). RECIST assumes that lung nodules routinely change size in a symmetric fashion. It is under these conditions that planar measurement would faithfully reflect the true extent of volume changes by a growing tumor nodule. Unfortunately, studies have revealed that tumors frequently do not grow symmetrically and that substantial variability exists in RECIST measurements (8). Increased precision is necessary as some studies have discovered that RECIST failed to accurately identify up to 33% of patients adequately responding to tyrosine kinase inhibitors (9). By not providing the correct assessment, clinicians may compromise the care of cancer patients. For example, in one case of an asymmetric nodule followed over time with serial CT scans, traditional measures classified the doubling time as benign, whereas 3D-based measurements revealed the doubling time to be consistent with malignancy (7). Concern with such inaccuracies has encouraged movement toward 3D imaging. The development of software like the ISP tool may facilitate the way clinicians integrate quantitative imaging into routine clinical management (10,11). The best example of an analogous situation involving the transition to 3D visualization tools exists in the architectural community where it took several decades to learn how to integrate the enormous amount of digital data generated using computer-aided building design software. The use of 3D digital tool transformed the field of architecture and demonstrates the potential contribution of managing digital data as a resource for shared planning and education locally or across the web. Interestingly, the original motivation for moving architectural planning from traditional blueprints to computer-aided design (CAD) formats was to improve the accuracy of the design documents. However, it was quickly appreciated that CAD planning facilitated collaborative interaction among participants in the design process in profound and unanticipated ways. The notable achievements of Frank Gehry illustrate how CAD emerged as a transformative tool in architecture. The Gehry
EVALUATION OF THE ISP TOOL
design team, which was a late adapter of 3D design, was initially characterized as CAD phobic (12); however, after many years of aversion, they rapidly assimilated process knowledge and soon emerged as radical innovators, producing a series of original (yet already iconic) structures. Key to their success was an understanding that the 3D design process was disruptive but in a different way for different stakeholders. The ability to visualize the entire integrated planned structured in scale allowed simultaneous tailored evolution of numerous construction processes, such as structural engineering, sound management, fire safety, and dry wall fabrication. This digital planning environment resulted in a new environment (or matrix) of innovation by allowing simultaneous interaction with the same digital data by separate specialists involved in the construction planning process. Each construction discipline solved its respective planning problem by interacting with the shared digital data that were constrained by the dynamic changes to the building plans added in real time by all the disciplines of building designers. This type of digital environment provided a level of integration, analysis, and coordination that had not been previously feasible. Furthermore, architecture is not the only field in which 3D planning has been transformational. Another example of this disruptive potential exists with the integration of 3D industrial fabrication into the progress in computer chip design which made possible a stacked rather than planar chip designs allowing significant improvement with heat dissipation (13). The lesson of the architectural experience with 3D innovation is that this type of digital planning and analysis capability is most relevant to multidisciplinary medical problem solving. Typically, in volumetric imaging applications, radiologists are focused on accurate and robust measurement of imaging findings to address a diagnostic question. A surgeon uses 3D imaging information functionally to guide therapeutic interventions. A neurosurgeon, for example, uses 3D imaging to assess luminal patency to assist in the selection of the proper interventional device for effective evacuation of the occluding clot from a vascular structure to limit the extent of damage from an evolving stroke. Similarly, 3D imaging has enabled a new functional classification of spinal scoliosis on the basis of the nature of torsional stress across spinous structures—a perspective that was not possible with the use of 2D spinal images (14). The pace of biomedical evolution may accelerate with the use of 3D data sets as communication/collaboration tools. This was the goal of the Optical Society in developing ISP to serve as such a reference visualization resource to facilitate research collaboration (4). ISP can serve as an infrastructure to systematically acquire and host libraries of medical images as a Web-accessible resource. In certain instances, libraries of image data sets may also include linked clinical outcome data from the individuals who donated the medical images. Such image/clinical outcome libraries constitute an invaluable resource for image software development and other types of research and education that could be scaled using hosted Web architecture ensuring that this resource would be economically accessible to the global medical community. 385
RINEWALT ET AL
Beyond clinical applications of CT, the underlying 3D rendering and related visualization approaches are being applied to basic studies of subcellular structures (15). The ISP environment may serve as a new tool to enable collaboration to scale the divide between basic and clinical science research. This ISP collaboration tool, which enables shared access to primary imaging data across all disciplines while providing a toolkit of reference for quantitative imaging techniques, may represent a new foundational resource for research convergence (16). From this pilot evaluation, ISP software application seems to be a useful tool for viewing and interpreting radiographic images and provides a strategic resource for serving as a collaboration tool to allow many groups to interact dynamically with the same digital (DICOM) imaging data set. As the ISP tool was found to have minimal (if any) diminution in accuracy compared to traditional methods and was favored by new users, we propose that increased use of this technology is warranted. Further analysis of user perspectives around the 3D innovation process may assist with catalyzing progress with imaging while also providing an opportunity to better understand and facilitate the process of innovation. ACKNOWLEDGMENT The authors wish to thank Ricardo Avila for his critical assistance and technical advice in the development of the interactive science publishing evaluation tool and Josephine Volgi, RN, for her gracious perseverance in administering all the evaluations. The authors thank Artit Jirapatnakul for developing the PDF vehicle to administer the survey tool in the defined sequence while allowing recovery of the evaluation responses. They also thank David Shersher for his help with identifying volunteers for this effort and the many Rush physicians, who volunteered their time to participate in the evaluation process. The authors appreciate the support for the research and development of this article by a contract with the Optical Society of America and the
386
Academic Radiology, Vol 22, No 3, March 2015
US Air Force. We appreciate the critical review of the article by Dr. Thomas Baer, Photonic Research Center at Stanford University.
REFERENCES 1. Mulshine JL, Avila R, Yankelevitz D, et al. Use of high resolution CT imaging data in lung cancer drug development: measuring progress. Oncology 2009; 23:434–438. 2. Buckler AJ, Mulshine JL, Gottlieb R, et al. The use of volumetric CT as an imaging biomarker in lung cancer. Acad Radiol 2010; 17:100–106. 3. Eisenhauer EA, Therasse P, Bogaerts J, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 2009; 45:228–247. 4. Mulshine JL, Baer TM, Avila RS. Introduction: imaging in diagnosis and treatment of lung cancer. Opt Express 2010; 18:15242–15243. 5. Aberle DR, Adams AM, Berg CD, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011; 365: 395–409. 6. Mulshine JL, Jablons DM. Volume CT for diagnostics work-up in a randomized lung cancer screening trial: learning from Nelson. N Engl J Med 2009; 361:2221–2229. 7. Yankelevitz DF, Reeves AP, Kostis WJ, et al. Small pulmonary nodules: volumetrically determined growth rates based on CT evaluation. Radiology 2000; 217:251–256. 8. Gavrielides MA, Kinnard LM, Myers KJ, et al. Noncalcified lung nodules: volumetric assessment with thoracic CT. Radiology 2009; 251:26–37. 9. Zhao B, Oxnard GR, Moskowitz CS, et al. A pilot study of volume measurement as a method of tumor response evaluation to aid biomarker development. Clin Cancer Res 2010; 16:4647–4653. 10. Revel M, Lefort C, Bissery A, et al. Pulmonary nodules: preliminary experience with three-dimensional evaluation. Radiology 2004; 231:459–466. 11. Krishnan K, Ibanez L, Turner WD, et al. An open-source toolkit for the volumetric measurement of CT lung lesions. Optics Express 2010; 18: 15256–15266. 12. Boland RJ, Lyytinen K, Yoo Y. Wakes of innovation in project networks: the case of digital 3-D representations in architecture, engineering, and construction. Organization Science 2007; 18:631–647. 13. Young AM, Koester SJ. 3D process technology considerations. In: Xie Y, Cong J, Sapatnekar S, eds. Three-dimensional integrated circuit design, integrated circuits and systems. New York, NY: Springer LLC, 2010. 14. Labelle H, Aubin CE, Jackson R, et al. Seeing the spine in 3D: how will it change what we do? J Pediatr Orthop 2011; 31:S37–S45. 15. Miao Q, Yu J, Rahn JR, et al. Dual-mode optical projection tomography microscope using gold nanorods and hematoxylin-stained cancer cells. Opt Lett 2010; 35:1037–1039. 16. Sharp PA, Langer R. Promoting convergence in biomedical science. Science 2011; 333:527.
