Multiview-based computer-aided detection scheme for breast masses.

In this study, we developed and tested a new multiview-based computer-aided detection (CAD) scheme that aims to maintain the same case-based sensitivity level as a single-image-based scheme while substantially increasing the number of masses being detected on both ipsilateral views. An image database of 450 four-view examinations (1800 images) was assembled. In this database, 250 cases depicted malignant masses, of which 236 masses were visible on both views and 14 masses were visible only on one view. First, we detected suspected mass regions depicted on each image in the database using a single-image-based CAD. For each identified region (with detection score > or = 0.55), we then identified a matching strip of interest on the ipsilateral view based on the projected distance to the nipple along the centerline. By lowering CAD operating threshold inside the matching strip, we searched for a region located inside the strip and paired it with the original region. A multifeature-based artificial neural network scored the likelihood of the paired "matched" regions representing true-positive masses. All single (unmatched) regions except for those either with very high detection scores (> or = 0.85) or those located near the chest wall that cannot be matched on the other view were discarded. The original single-image-based CAD scheme detected 186 masses (74.4% case-based sensitivity) and 593 false-positive regions. Of the 186 identified masses, 91 were detected on two views (48.9%) and 95 were detected only on one view (51.1%). Of the false-positive detections, 54 were paired on the ipsilateral view inside the corresponding matching strips and the remaining 485 were not, which represented 539 case-based false-positive detections (0.3 per image). Applying the multiview-based CAD scheme, the same case-based sensitivity was maintained while cueing 169 of 186 masses (90.9%) on both views and at the same time reducing the case-based false-positive detection rate by 23.7% (from 539 to 411). The study demonstrated that the new multiview-based CAD scheme could substantially increase the number of masses being cued on two ipsilateral views while reducing the case-based false-positive detection rate.

A multisite telemammography system for remote management of screening mammography: an assessment of technical, operational, and clinical issues.

OBJECTIVE: This paper describes a high-quality, multisite telemammography system to enable "almost real-time" remote patient management while the patient remains in the clinic. One goal is to reduce the number of women who would physically need to return to the clinic for additional imaging procedures (termed "recall") to supplement "routine" imaging of screening mammography. MATERIALS AND METHODS: Mammography films from current and prior (when available) examinations are digitized at three remote sites and transmitted along with other pertinent information across low-level communication systems to the central site. Images are automatically cropped, wavelet compressed, and encrypted prior to transmission to the central site. At the central site, radiologists review and rate examinations on a high-resolution workstation that displays the images, computer-assisted detection results, and the technologist's communication. Intersite communication is provided instantly via a messaging "chat" window. RESULTS: The technologists recommended additional procedures at 2.7 times the actual clinical recall rate for the same cases. Using the telemammography system during a series of "off-line" clinically simulated studies, radiologists recommended additional procedures at 1.3 times the actual clinical recall rate. Percent agreement and kappa between the study and actual clinical interpretations were 66.1% and 0.315, respectively. For every physical recall potentially avoided using the telemammography system, approximately one presumed "unnecessary" imaging procedure was recommended. CONCLUSION: Remote patient management can reduce the number of women recalled by as much as 50% without performing an unreasonable number of presumed "unnecessary" procedures.

The effect of image display size on observer performance an assessment of variance components.

RATIONALE AND OBJECTIVE: Our goal was to investigate the effect of the displayed image size on variance components during the performance of an observer performance study to detect masses on abdominal computed tomography (CT) examinations. MATERIALS AND METHODS: A previously performed receiver operating characteristic (ROC) study with eight observers to detect abdominal masses on 166 CT examinations was reanalyzed to assess variance components when comparing two similar modes with displayed image sizes varying by a factor of 2. Case, mode, and reader-related variance components were estimated for the group of eight observers and subsets of readers after excluding each of the participants. RESULTS: There was no significant difference in the average area under the ROC curves between the two modes using the two image sizes (P > .05). Reader and reader-by-case variability were substantially larger for the mode displaying enlarged images for the group and all subsets formed by excluding a single reader. Reader variability was affected by one observer who actually performed better with the enlarged images. CONCLUSION: Sequential viewing of enlarged CT images for the detection of abdominal masses did not improve performance and increased reader variability.

Head-mounted versus remote eye tracking of radiologists searching for breast cancer: a comparison.

PURPOSE: We compared performance and visual search parameters of radiologists detecting masses on mammograms by using both a head-mounted (HDMT) and a remote (REM) eye tracker. MATERIALS AND METHODS: Five experienced radiologists read twice a case set of 20 one-view (medial-lateral oblique) mammograms, of which 12 contained a malignant mass and eight were lesion-free. For each observer, one trial used an HDMT eye-tracking system and the other used an REM system. Trials were separated on average by 2 months. Time to hit the location of the mass, dwell, and number of fixations in the location of the mass were measured. The same parameters were measured on a per-trial basis to determine whether there were memory effects from the previous trial. RESULTS: Dwell times in the location of true-positive, false-positive, and false-negative results were significantly shorter (P < .05) using the HDMT (median, 0.395 seconds) than REM (median, 0.482 seconds) systems, but the number of fixations in the location of the response was smaller using the REM system (median, 4.33 versus 5.0 for the HDMT). The observed differences did not seem to be caused by a memory effect. In addition, the relative lack of head mobility using the REM system caused observers to report neck strain. CONCLUSION: Overall, radiologists' visual search behavior was very similar using both types of eye-tracking device. However, because the REM system did not contain a magnetic head tracker, radiologists were allowed very limited head movements when using it, which made them uncomfortable during the experiment.

Variability in observer performance studies experimental observations.

RATIONALE AND OBJECTIVES: The aim of the study is to assess variance components in observer performance studies and the possible impact on study results and conclusions. MATERIALS AND METHODS: Two previously performed retrospective receiver operating characteristic-type observer performance studies to evaluate the performance of seven radiologists in detecting interstitial disease on conventional posteroanterior chest films and nine radiologists in detecting interstitial disease on a high-resolution workstation were reanalyzed by using the Beiden, Wagner, and Campbell nine-component model to estimate the different variance components. We estimated case-, reader-, and mode-related components of the variance for the group as a whole and after excluding (round robin) each reader. Overall variance was evaluated, and the effect of individual readers on overall study conclusions was assessed. RESULTS: Overall results and conclusions of the reanalysis agreed with the original one in that, as a group, radiologists performed significantly better when using conventional films (P < .05) in both studies. Reader variability was large compared with all other components, and in one study, it was substantially larger for the workstation reading mode. Reader variability was affected substantially by one observer in each study, and in one study, reader-by-mode variability was affected by another reader who performed better on the workstation. CONCLUSION: Estimates of variance components can shed light on the appropriateness of study design, as well as the sensitivity of results to the inclusion (or exclusion) of individual observers.

Effects of lesion conspicuity on visual search in mammogram reading.

RATIONALE AND OBJECTIVES: The goal of mammography screening is to detect breast cancer at early stages, but because of the complexity of the breast parenchyma and the variability of signs of the disease, many cancers go unreported when initially visible on the mammogram. We compared the visual search strategy used by experienced mammographers in a case set where they examined both the mammogram in which a malignant mass was discovered at screening mammography and the most recent prior mammogram. MATERIALS AND METHODS: Four experienced mammographers participated in this experiment. They read a case set of 20 two-view mammograms, of which 15 contained a malignant mass and 5 were lesion-free, in two trials. For each of the cancer cases, two versions were shown to the observers: the one in which the cancer was reported in the clinical practice, called the "current" mammograms, and the most recent prior. Each trial had a balanced mix of current and prior mammograms. In addition, the same set of lesion-free cases was shown to the observers in both trials. The eye movements of the observers were tracked, and visual search parameters such as time to hit the location of the malignant mass, dwell, and mean pupil size in the location of the cancer were collected. Statistical analyses were used to determine whether there were differences between the current and prior mammograms. RESULTS: A total of 66% of the malignant masses in the current mammograms and 57% in the priors attracted some amount of visual attention. From these, 71% yielded a report on the current mammograms, but only 40% on the priors. In the cases where the observer saw the malignant mass, they did so within 2 seconds of image display, regardless of whether the mammogram was current or prior. CONCLUSION: Most unreported malignant masses attracted some amount of visual attention, but it was in the processing of the information extracted in the location of the lesion that most errors occurred. In our experiment, approximately 70% of the total time used by the observers for visual scan of the cases was spent gathering information to corroborate the hypothesis already formed by the radiologist.

Performance and reproducibility of a computerized mass detection scheme for digitized mammography using rotated and resampled images: an assessment.

OBJECTIVE: Our objective was to compare the performance and reproducibility of a computer-aided detection (CAD) scheme that uses multiple rotated and resampled images with an in-house-developed CAD scheme (single-image-based) and a commercial CAD product in detecting masses depicted on digitized mammograms. MATERIALS AND METHODS: Ninety-two film mammograms (acquired from 23 patients) were selected. Forty-four mass regions associated with malignancy were visually identified. A commercial CAD system was used to scan and process each image four times, for a total of 368 digitized images depicting 176 mass regions. Images were processed using two CAD schemes developed in our laboratory. One uses the detection results generated from a single image, and the other averages five detection scores generated after processing the originally digitized image and four slightly rotated and resampled images. A region-based analysis was used to compare reproducibility and performance levels among the two in-house schemes and the commercial system. RESULTS: The commercial system detected a total of 98 mass regions (55.7% sensitivity) and 136 false-positive regions (an average of 0.37 per image). Among the detected mass regions, 76 represented 19 regions that were detected on all four scans and 22 represented 10 regions that were not fully reproducible. Eighty-eight false-positive detections represented 22 reproducible detections on all four scans. Our single-image-based scheme identified 87 mass regions and 160 false-positive regions. Seventeen mass regions and 28 false-positive regions were detected on all four scans. The multiple-image-based scheme identified 98 mass regions and 132 false-positive regions. Twenty-three mass regions were detected on all four scans. One hundred twelve of the 132 false-positive regions represented 28 reproducible detections. CONCLUSION: Averaging detection scores from multiple rotated and resampled images generated from a single digitization of a film can reduce variations in detection scores. Our multiple-image-based scheme improved both performance and reproducibility over the single-image-based scheme. The multiple-image-based scheme yielded an overall performance comparable to that of the commercial system but with improved reproducibility.

A comparison of two data analyses from two observer performance studies using Jackknife ROC and JAFROC.

The authors compared two methodological approaches, Jackknife ROC and JAFROC, in analyzing data ascertained during FROC (free-response receiver operating characteristics) type studies. Observer rating data obtained from two observer performance studies were analyzed. During the first study, seven radiologists interpreted 120 mammography examinations depicting 57 masses under five different conditions with and without the results of computer-aided detection (CAD). In the second study, eight radiologists interpreted 110 examinations depicting 51 masses under six different display conditions with and without CAD results. Readers rated the detection task in a FROC type response. Jackknife ROC (using the software of LABMRMC with the highest rating per case) and JAFROC were used to compute differences, if any, in summary performance levels among all reading modes in each study as well as for all paired data sets. The results of the different analytical approaches are compared. The overall results for all modes were significantly different for the first study (p < 0.05) and not significant (p > 0.05) for the second study using either analytical approach. In the first study, the performance levels represented by three paired data sets were significantly different (p < 0.05) when computed using LABMRMC and four pairs were significantly different (p < 0.05) using JAFROC. In eight of ten pairs, JAFROC produced lower p values than LABMRMC. In the second study, LABMRMC showed no significant differences for any paired data sets and JAFROC showed a significant difference for one pair. In 15 of 16 pairs, p values computed by JAFROC were lower than those computed by LABMRMC.