Radiologists are critically interested in promoting best practices in medical imaging, and to that end, they are actively developing tools that will optimize terminology and reporting practices in radiology. The RadLex® vocabulary, developed by the Radiological Society of North America (RSNA), is intended to create a unifying source for the terminology that is used to describe medical imaging. The RSNA Reporting Initiative has developed a library of reporting templates to integrate reusable knowledge, or meaning, into the clinical reporting process. This report presents the initial analysis of the intersection of these two major efforts. From 70 published radiology reporting templates, we extracted the names of 6,489 reporting elements. These terms were reviewed in conjunction with the RadLex vocabulary and classified as an exact match, a partial match, or unmatched. Of 2,509 unique terms, 1,017 terms (41%) matched exactly to RadLex terms, 660 (26%) were partial matches, and 832 reporting terms (33%) were unmatched to RadLex. There is significant overlap between the terms used in the structured reporting templates and RadLex. The unmatched terms were analyzed using the multidimensional scaling (MDS) visualization technique to reveal semantic relationships among them. The co-occurrence analysis with the MDS visualization technique provided a semantic overview of the investigated reporting terms and gave a metric to determine the strength of association among these terms.
Radiology; Structured reporting; Reporting templates; Standardized terminology; RadLex; Mapping; Visualization; Multidimensional scaling
Imaging centers nationwide are seeking innovative means to record and monitor computed tomography (CT)-related radiation dose in light of multiple instances of patient overexposure to medical radiation. As a solution, we have developed RADIANCE, an automated pipeline for extraction, archival, and reporting of CT-related dose parameters. Estimation of whole-body effective dose from CT dose length product (DLP)—an indirect estimate of radiation dose—requires anatomy-specific conversion factors that cannot be applied to total DLP, but instead necessitate individual anatomy-based DLPs. A challenge exists because the total DLP reported on a dose sheet often includes multiple separate examinations (e.g., chest CT followed by abdominopelvic CT). Furthermore, the individual reported series DLPs may not be clearly or consistently labeled. For example, “arterial” could refer to the arterial phase of the triple liver CT or the arterial phase of a CT angiogram. To address this problem, we have designed an intelligent algorithm to parse dose sheets for multi-series CT examinations and correctly separate the total DLP into its anatomic components. The algorithm uses information from the departmental PACS to determine how many distinct CT examinations were concurrently performed. Then, it matches the number of distinct accession numbers to the series that were acquired and anatomically matches individual series DLPs to their appropriate CT examinations. This algorithm allows for more accurate dose analytics, but there remain instances where automatic sorting is not feasible. To ultimately improve radiology patient care, we must standardize series names and exam names to unequivocally sort exams by anatomy and correctly estimate whole-body effective dose.
RADIANCE; Computed tomography; Dose monitoring; CT series separation; Radiation dose; Data extraction; Databases
The productivity gains, diagnostic benefit, and enhanced data availability to clinicians enabled by picture archiving and communication systems (PACS) are no longer in doubt. However, commercial PACS offerings are often extremely expensive initially and require ongoing support contracts with vendors to maintain them. Recently, several open-source offerings have become available that put PACS within reach of more users. However, they can be resource-intensive to install and assure that they have room for future growth—both for computational and storage capacity. An alternate approach, which we describe herein, is to use PACS built on virtual machines which can be moved from smaller to larger hardware as needed in a just-in-time manner. This leverages the cost benefits of Moore's Law for both storage and compute costs. We describe the approach and current results in this paper.
PACS; Open-source; Software; Digital subtraction angiography
It is difficult to detect sentinel lymph nodes (SLNs) around an injection point of radiopharmaceuticals mapped in lymphoscintigrams. The purpose of this study was to develop a computer-aided detection (CAD) scheme for SLNs by a subtraction technique using the symmetrical property in the mapped injection point. Our database consisted of 78 lymphoscintigrams with 86 SLNs. In our CAD scheme, the mapped injection point of radiopharmaceuticals was first segmented from the lymphoscintigram using a gray-level thresholding technique. Lymphoscintigram was then divided into four regions by vertical and horizontal straight lines through the center of the segmented injection point. One of the four divided regions was defined as the target region. The correlation coefficients based on pixel values were calculated between the target region and each of the other three regions. The region with the highest correlation coefficient among three regions was selected as the similar region to the target region. The values of pixels on the target region were subtracted by the values of the corresponding pixels on the similar region. This procedure was repeated until every divided region had been used as target region. SLNs were segmented by applying a gray-level thresholding technique to the subtracted image. With our CAD scheme, sensitivity and the number of false positives were 95.3% (82/86) and 2.51 per image, respectively. Our CAD scheme achieved a high level of detection accuracy, and would have a great potential in assisting physicians to detect SLNs in lymphoscintigrams.
Computer-aided detection; Lymphoscintigram; Sentinel lymph node; Image subtraction
Radiology conferences enable participants the opportunity to ask experts questions through question and answer (Q and A) sessions or individually. Given the time limitations and intimidating circumstances, we incorporated conference text messaging (confexting) as a method of increasing interactivity between the audience and speakers. During a 5-day radiology conference, text messaging was utilized for anonymous interactivity between the audience and speakers during Q and A sessions. There were 324 text messages; 76 of these were either follow-up statements or questions related to earlier text messages. Forty-two questions were submitted via paper notes. There was a general trend of an increasing number of text messages and a decreasing number of paper notes. The anonymous text messaging system was found to be an effective method for interactivity between the audience and the speakers. The questions and answers could be presented in a PowerPoint format at the formal Q and A sessions. Questions texted to the authors during their talks could be immediately answered or addressed in subsequent talks. Although difficult for some individuals to embrace technology, confexting allows for interactivity and prompts discussion. Confexting is an effective method for interactivity between the audience and speakers not previously utilized in a conference setting. The anonymity and asynchronous communication enable conference participants to submit more questions than in the traditional setting. The speakers may be able to explain more thoroughly difficult concepts more thoroughly with additional slides at Q and A sessions or may immediately answer texted questions during their talks.
Teaching; Continuing medical education; Computer hardware; Communication; Education; Medical; Experiential; Imaging informatics; User interface
This study examined whether radiology report format influences reading time and comprehension of information. Three reports were reformatted to conventional free text, structured text organized by organ system, and hierarchical structured text organized by clinical significance. Five attending radiologists, five radiology residents, five internal medicine attendings, and five internal medicine residents read the reports and answered a series of questions about them. Reading was timed and participants reported reading preferences. For reading time, there was no significant effect for format, but there was for attending versus resident, and radiology versus internal medicine. For percent correct scores, there was no significant effect for report format or for attending versus resident, but there was for radiology versus internal medicine with the radiologists scoring better overall. Report format does not appear to impact viewing time or percent correct answers, but there are differences in both for specialty and level of experience. There were also differences between the four groups of participants with respect to what they focus on in a radiology report and how they read reports (skim versus read in detail). There may not be a “one-size-fits-all” radiology report format as individual preferences differ widely.
Radiology reporting; Workflow; Communication
Data sharing is increasingly recognized as critical to cross-disciplinary research and to assuring scientific validity. Despite National Institutes of Health and National Science Foundation policies encouraging data sharing by grantees, little data sharing of clinical data has in fact occurred. A principal reason often given is the potential of inadvertent violation of the Health Insurance Portability and Accountability Act privacy regulations. While regulations specify the components of private health information that should be protected, there are no commonly accepted methods to de-identify clinical data objects such as images. This leads institutions to take conservative risk-averse positions on data sharing. In imaging trials, where images are coded according to the Digital Imaging and Communications in Medicine (DICOM) standard, the complexity of the data objects and the flexibility of the DICOM standard have made it especially difficult to meet privacy protection objectives. The recent release of DICOM Supplement 142 on image de-identification has removed much of this impediment. This article describes the development of an open-source software suite that implements DICOM Supplement 142 as part of the National Biomedical Imaging Archive (NBIA). It also describes the lessons learned by the authors as NBIA has acquired more than 20 image collections encompassing over 30 million images.
Data sharing; De-identification; Anonymization; Cross-disciplinary research; Open access; Open source; DICOM; Supplement 142; Image archive; HIPAA; PHI; Common rule
Cardiovascular disease is the leading cause of global mortality, yet its early detection remains a vexing problem of modern medicine. Although the computed tomography (CT) calcium score predicts cardiovascular risk, relatively high cost ($250–400) and radiation dose (1–3 mSv) limit its universal utility as a screening tool. Dual-energy digital subtraction radiography (DE; <$60, 0.07 mSv) enables detection of calcified structures with high sensitivity. In this pilot study, we examined DE radiography’s ability to quantify coronary artery calcification (CAC). We identified 25 patients who underwent non-contrast CT and DE chest imaging performed within 12 months using documented CAC as the major inclusion criteria. A DE calcium score was developed based on pixel intensity multiplied by the area of the calcified plaque. DE scores were plotted against CT scores. Subsequently, a validation cohort of 14 additional patients was independently evaluated to confirm the accuracy and precision of CAC quantification, yielding a total of 39 subjects. Among all subjects (n = 39), the DE score demonstrated a correlation coefficient of 0.87 (p < 0.0001) when compared with the CT score. For the 13 patients with CT scores of <400, the correlation coefficient was −0.26. For the 26 patients with CT scores of ≥400, the correlation coefficient yielded 0.86. This pilot study demonstrates the feasibility of DE radiography to identify patients at the highest cardiovascular risk. DE radiography’s accuracy at lower scores remains unclear. Further evaluation of DE radiography as an inexpensive and low-radiation imaging tool to diagnose cardiovascular disease appears warranted.
Calcification detection; Cardiac imaging; Chest CT; Chest radiographs; Computed tomography; Coronary arteries; Coronary calcifications; Coronary disease; Digital radiography; Digital subtraction radiography; Dual-energy subtraction; Radiography; Dual-energy scanned projection; ROC-based analysis
Successful adoption of new technology development can be accentuated by learning and applying the scientific principles of innovation diffusion. This is of particular importance to areas within the medical imaging practice which have lagged in innovation; perhaps, the most notable of which is reporting which has remained relatively stagnant for over a century. While the theoretical advantages of structured reporting have been well documented throughout the medical imaging community, adoption to date has been tepid and largely relegated to the academic and breast imaging communities. Widespread adoption will likely require an alternative approach to innovation, which addresses the heterogeneity and diversity of the practicing radiologist community along with the ever-changing expectations in service delivery. The challenges and strategies for reporting innovation and adoption are discussed, with the goal of adapting and customizing new technology to the preferences and needs of individual end-users.
Innovation adoption; Structured reporting; Medical imaging
Because of a much higher dynamic range of flat panel detectors, patient dose can vary without change of image quality being perceived by radiologists. This condition makes optimization (OT) of radiation protection undergoing digital radiography (DR) more complex, while a chance to reduced patient dose also exists. In this study, we evaluated the difference of patient radiation and image rejection before and after OT to identify if it is necessary to carry out an OT procedure in a routine task with DR. The study consisted of a measurement of the dose area product (DAP) and entrance surface dose (ESD) received by a reference group of patients for eight common radiographic procedures using the DR system before and after OT. Meanwhile image rejection data during two 2-month periods were collected and sorted according to reason. For every radiographic procedure, t tests showed significant difference in average ESD and DAP before and after OT (p < 0.005). The ESDs from most examinations before OT were three times higher than that after OT. For DAPs, the difference is more significant. Image rejection rate after OT is significantly lower than that before OT (χ2 = 36.5, p < 0.005). The substantial reductions of dose after OT resulted from appropriate mAs and exposure field. For DR patient dose, less than recommended diagnostic reference level can meet quality criteria and clinic diagnosis.
Optimization; Digital radiography; Radiation dose; Diagnostic image quality; Exposure index
The staggering number of images acquired by modern modalities requires new approaches for medical data transmission. There have been several attempts to improve data transmission time between medical imaging systems. These attempts were mostly based on compression. Although the compression methods can help in many cases, they are sometimes ineffectual in high-speed networks. This paper introduces parallelism to provide an effective method of medical data transmission over both local area network (LAN) and wide area network (WAN). It is based on the Digital Imaging and Communications in Medicine (DICOM) protocol and uses parallel TCP connections in storage services within the protocol. Using the proposed interface in our method, current medical imaging applications can take advantage of parallelism without any modification. Experimental results show a speedup of about 1.3 to 1.5 for CT images and relatively high speedup of about 2.2 to 3.5 times for magnetic resonance (MR) images over LAN. The transmission time is improved drastically over WAN. The speedup is about 16.1 for CT images and about 5.6 to 11.5 for MR images.
DICOM; Telemedicine; Parallelism
Imaging signs form an important part of the language of radiology, but are not represented in established lexicons. We sought to incorporate imaging signs into RSNA's RadLex® ontology of radiology terms. Names of imaging signs and their definitions were culled from books, journal articles, dictionaries, and biomedical web sites. Imaging signs were added into RadLex as subclasses of the term “imaging sign,” which was defined in RadLex as a subclass of “imaging observation.” A total of 743 unique imaging signs were added to RadLex with their 392 synonyms to yield a total of 1,135 new terms. All included definitions and related RadLex terms, including imaging modality, anatomy, and disorder, when appropriate. The information will allow RadLex users to identify imaging signs by modality (e.g., ultrasound signs) and to find all signs related to specific pathophysiology. The addition of imaging signs to RadLex augments its use to index the radiology literature, create and interpret clinical radiology reports, and retrieve relevant cases and images.
Knowledge representation; Information storage and retrieval; Image retrieval; RadLex; Imaging signs; Ontology
Optimization of brightness distribution in the template used for detection of cancerous masses in mammograms by means of correlation coefficient is presented. This optimization is performed by the evolutionary algorithm using an auxiliary mass classifier. Brightness along the radius of the circularly symmetric template is coded indirectly by its second derivative. The fitness function is defined as the area under curve (AUC) of the receiver operating characteristic (ROC) for the mass classifier. The ROC and AUC are obtained for a teaching set of regions of interest (ROIs), for which it is known whether a ROI is true-positive (TP) or false-positive (F). The teaching set is obtained by running the mass detector using a template with a predetermined brightness. Subsequently, the evolutionary algorithm optimizes the template by classifying masses in the teaching set. The optimal template (OT) can be used for detection of masses in mammograms with unknown ROIs. The approach was tested on the training and testing sets of the Digital Database for Screening Mammography (DDSM). The free-response receiver operating characteristic (FROC) obtained with the new mass detector seems superior to the FROC for the hemispherical template (HT). Exemplary results are the following: in the case of the training set in the DDSM, the true-positive fraction (TPF) = 0.82 for the OT and 0.79 for the HT; in the case of the testing set, TPF = 0.79 for the OT and 0.72 for the HT. These values were obtained for disease cases, and the false-positive per image (FPI) = 2.
Cancer detection; Image analysis; Mammography; Breast
In this paper, we present an effective method to determine the reference point of symphysis pubis (SP) in an axial stack of CT images to facilitate image registration for pelvic cancer treatment. In order to reduce the computational time, the proposed method consists of two detection parts, the coarse detector, and the fine detector. The detectors check each image patch whether it contains the characteristic structure of SP. The coarse detector roughly determines the location of the reference point of SP using three types of information, which are the location and intensity of an image patch, the SP appearance, and the geometrical structure of SP. The fine detector examines around the location found by the coarse detection to refine the location of the reference point of SP. In the experiment, the average location error of the propose method was 2.23 mm, which was about the side length of two pixels. Considering that the average location error by a radiologist is 0.77 mm, the proposed method finds the reference point quite accurately. Since it takes about 10 s to locate the reference point from a stack of CT images, it is fast enough to use in real time to facilitate image registration of CT images for pelvic cancer treatment.
Computer-aided diagnosis (CAD); Image registration; Pattern recognition; Computed tomography; Symphysis pubis; Haar-like features; Biased discriminant analysis
We have developed a method to quantify the shape of liver lesions in CT images and to evaluate its performance for retrieval of images with similarly-shaped lesions. We employed a machine learning method to combine several shape descriptors and defined similarity measures for a pair of shapes as a weighted combination of distances calculated based on each feature. We created a dataset of 144 simulated shapes and established several reference standards for similarity and computed the optimal weights so that the retrieval result agrees best with the reference standard. Then we evaluated our method on a clinical database consisting of 79 portal-venous-phase CT liver images, where we derived a reference standard of similarity from radiologists’ visual evaluation. Normalized Discounted Cumulative Gain (NDCG) was calculated to compare this ordering with the expected ordering based on the reference standard. For the simulated lesions, the mean NDCG values ranged from 91% to 100%, indicating that our methods for combining features were very accurate in representing true similarity. For the clinical images, the mean NDCG values were still around 90%, suggesting a strong correlation between the computed similarity and the independent similarity reference derived the radiologists.
Image retrieval; Image analysis; Image processing
A common teleradiology practice is digitizing films. The costs of specialized digitizers are very high, that is why there is a trend to use conventional scanners and digital cameras. Statistical clinical studies are required to determine the accuracy of these devices, which are very difficult to carry out. The purpose of this study was to compare three capture devices in terms of their capacity to detect several image characteristics. Spatial resolution, contrast, gray levels, and geometric deformation were compared for a specialized digitizer ICR (US$ 15,000), a conventional scanner UMAX (US$ 1,800), and a digital camera LUMIX (US$ 450, but require an additional support system and a light box for about US$ 400). Test patterns printed in films were used. The results detected gray levels lower than real values for all three devices; acceptable contrast and low geometric deformation with three devices. All three devices are appropriate solutions, but a digital camera requires more operator training and more settings.
Teleradiology; Diagnostic image quality; Image acquisition; Image quality; PACS; Image viewer; Film digitizer; X-ray digital capture
Scoliosis is a 3-D deformity of spinal column, characterized by both lateral curvature and vertebral rotation. The disease can be caused by congenital, developmental, or degenerative problems; but most cases of scoliosis actually have no known cause, and this is known as idiopathic scoliosis. Vertebral rotation has become increasingly prominent in the study of scoliosis and the most deformed vertebra is named as apical vertebra. Apical vertebral deformity demonstrates significance in both preoperative and postoperative assessment, providing better appreciation of the impact of bracing or surgical interventions. Precise measurement of apical vertebral rotation in terms of grading is most valuable for the determination of reference value in normal and pathological conditions for better understanding of scoliosis. Routine quantitative evaluation of vertebral rotation is difficult and error prone due to limitations of observer characteristic and specific imaging property. This paper proposes automatic identification of the apical vertebra and its parameter that depends on the objective criteria of measurement using active contour models. The proposed technique is more accurate and is a reliable measurement compared to manual and computer-assisted system.
Vertebral rotation; Scoliosis; Nash–Moe; Pedicle displacement
A patient has an imaging study performed at one facility and has the study exported to portable media. Later, the patient takes the media to a different institution. The study on that media may need to be imported into that new institution’s imaging system. This would be done to avoid a repeat examination, or so that the study can be on file for reference purposes. Importing prior studies is best performed by creating a new order on the institution’s imaging system and then associating the DICOM objects from the prior study with it. In this way the prior study is actually inserted into the imaging system’s electronic health record (EHR) and is properly indexed so that it can be identified and later retrieved as needed. In the past at the Department of Veterans Affairs (VA), importing prior DICOM studies into the VA systems had been a very slow labor-intensive process that took anywhere from 10 to 30 min to import a single study. We have developed a new DICOM Importer application that reduces the manual effort to import a prior study to less than a minute. We have redesigned and automated the process to make it much more efficient for the user. The Importer also handles contract examinations that are ordered by the VA and performed at outside imaging facilities, with similar time savings. This work is important because is addresses one of the major unsolved problems with import reconciliation workflow: how to efficiently handle the importing of prior studies.
DICOM studies; Imaging system; Importation; Hospital Information Systems (HIS); Image Acquisition; Digital Image Management; Digital Imaging and Communications in Medicine (DICOM); Integrating Healthcare Enterprise (IHE); Enterprise PACS; PACS DICOM IHE Conformance; Workflow reengineering
The aims of this study were to (1) investigate the repeatability of measured volumes using the atlas-based method in each area of the brain, and (2) validate our hypothesis that the repeatability of the measured volumes with the atlas-based method was improved by using smoothed images. T1-weighted magnetic resonance images were obtained in five healthy subjects using the 1.5-T scanner. We used Statistical Parametric Mapping 5 and WFU PickAtlas software (theory of the Talairach brain atlas). Volumes inside region-of-interest (ROI) were measured in ten sets (five subjects × right and left) on six ROIs, respectively. One set comprises five images (one subject × five 3D-T1WIs). The percentage change was defined as [100 × (measured volume–mean volume in each set)/mean volume in each set)]. As a result, the average percentage changes using non-smoothed image on each ROI were as follows: gray matter, 0.482%; white matter, 0.375%; cerebrospinal fluid images, 0.731%; hippocampus, 0.864%; orbital gyrus, 1.692%; cerebellum posterior lobe, 0.854%. Using smoothed images with large FWHM resulted in improved repeatability on orbital gyrus. This is the first report of repeatability in each brain structure and improved repeatability with smoothed images using the atlas-based method.
Atlas-based method; Brain volumetry; Magnetic resonance imaging; Repeatability; WFU PickAtlas; Brain imaging; Brain mapping; Brain morphology; Clinical application; Computer analysis; Image analysis
The objective of this study was to implement and evaluate the performance of a biplane correlation imaging (BCI) technique aimed to reduce the effect of anatomic noise and improve the detection of lung nodules in chest radiographs. Seventy-one low-dose posterior–anterior images were acquired from an anthropomorphic chest phantom with 0.28° angular separations over a range of ±10° along the vertical axis within an 11 s interval. Similar data were acquired from 19 human subjects with institutional review board approval and informed consent. The data were incorporated into a computer-aided detection (CAD) algorithm in which suspect lesions were identified by examining the geometrical correlation of the detected signals that remained relatively constant against variable anatomic backgrounds. The data were analyzed to determine the effect of angular separation, and the overall sensitivity and false-positives for lung nodule detection. The best performance was achieved for angular separations of the projection pairs greater than 5°. Within that range, the technique provided an order of magnitude decrease in the number of false-positive reports when compared with CAD analysis of single-view images. Overall, the technique yielded ~1.1 false-positive per patient with an average sensitivity of 75%. The results indicated that the incorporation of angular information can offer a reduction in the number of false-positives without a notable reduction in sensitivity. The findings suggest that the BCI technique has the potential for clinical implementation as a cost-effective technique to improve the detection of subtle lung nodules with lowered rate of false-positives.
Computer-aided detection; Biplane correlation imaging; Lung nodules; Chest radiography
Optimization and standardization of radiographic procedures in a health region minimizes patient exposure while producing diagnostic images. This report highlights the dose variation in common computed radiography (CR) examinations throughout a large health region. The RadChex cassette was used to measure the radiation exposure at the table or wall bucky in 20 CR rooms, in seven hospitals, using CR technology from two vendors. Exposures were made to simulate patient exposure (21 cm polymethyl methacrylate) under standard conditions for each bucky: 81 kVp at 100 cm for anteroposterior abdomen table bucky exposures (180 cm for posteroanterior chest wall bucky exposures), using the left, the right, or the center automatic exposure control (AEC) cells. Protocol settings were recorded. An average of 37% variation was found between AEC chambers, with a range between 4% and 137%. A 60% difference in dose was discovered between manufacturers, which was the result of the manufacture’s image processing algorithm and subsequently corrected via software updates. Finally, standardizing AEC cell selection during common chest examinations could reduce patient dose by up to 30%. In a large health region, variation in exam protocols can occur, leading to unnecessary patient dose from the same type of examination. Quality control programs must monitor exam protocols and AEC chamber calibration in CR to ensure consistent, minimal, patient dose, regardless of hospital or CR vendor. Furthermore, this report highlights the need for communication between radiologists, technologists, medical physicist, service engineers, and manufacturers required to optimize CR protocols.
Patient dose; Computed radiography; Radiation protection; Radiography; Quality control
Imaging modality can aid retrieval of medical images for clinical practice, research, and education. We evaluated whether an ensemble classifier could outperform its constituent individual classifiers in determining the modality of figures from radiology journals. Seventeen automated classifiers analyzed 77,495 images from two radiology journals. Each classifier assigned one of eight imaging modalities—computed tomography, graphic, magnetic resonance imaging, nuclear medicine, positron emission tomography, photograph, ultrasound, or radiograph—to each image based on visual and/or textual information. Three physicians determined the modality of 5,000 randomly selected images as a reference standard. A “Simple Vote” ensemble classifier assigned each image to the modality that received the greatest number of individual classifiers’ votes. A “Weighted Vote” classifier weighted each individual classifier’s vote based on performance over a training set. For each image, this classifier’s output was the imaging modality that received the greatest weighted vote score. We measured precision, recall, and F score (the harmonic mean of precision and recall) for each classifier. Individual classifiers’ F scores ranged from 0.184 to 0.892. The simple vote and weighted vote classifiers correctly assigned 4,565 images (F score, 0.913; 95% confidence interval, 0.905–0.921) and 4,672 images (F score, 0.934; 95% confidence interval, 0.927–0.941), respectively. The weighted vote classifier performed significantly better than all individual classifiers. An ensemble classifier correctly determined the imaging modality of 93% of figures in our sample. The imaging modality of figures published in radiology journals can be determined with high accuracy, which will improve systems for image retrieval.
Computer vision; Content-based image retrieval; Digital libraries; Image analysis; Image retrieval; Classification; Data mining