Search PubMed⌕ Search

Biomedical subjects

N A Obuchowski

Publications and source records attributed to N A Obuchowski.

At least 19 recordsLinked to original sources

Assessing physicians' accuracy in diagnosing paediatric patients with acute abdominal pain: measuring accuracy for multiple diseases.

In evaluating the accuracy of diagnostic tests there are many situations where the true disease status can be one of multiple disorders. For example, when paediatric patients suffer acute abdominal pain, the underlying diagnosis could be appendicitis, intestinal obstruction, gastroenteritis, urinary tract infection, etc. In this paper we describe a format for collecting data for the case of multiple truth states. This new format allows one to make pairwise comparisons of the diagnostic accuracy between all of the different truth states. We propose a summary measure of accuracy which is a weighted average of the pairwise estimates of accuracy. Estimators are derived for the variance and covariance of the estimated summary accuracy. The small sample properties of the estimators are evaluated in a Monte Carlo simulation study. The new data collection format and summary measure were used in a paediatric acute abdominal pain study; data from this study are used to illustrate the methods.

Abdomen, Acute↗

Can electronic medical images replace hard-copy film? Defining and testing the equivalence of diagnostic tests.

Electronic medical images are an efficient and convenient format in which to display, store and transmit radiographic information. Before electronic images can be used routinely to screen and diagnose patients, however, it must be shown that readers have the same diagnostic performance with this new format as traditional hard-copy film. Currently, there exist no suitable definitions of diagnostic equivalence. In this paper we propose two criteria for diagnostic equivalence. The first criterion ('population equivalence') considers the variability between and within readers, as well as the mean reader performance. This criterion is useful for most applications. The second criterion ('individual equivalence') involves a comparison of the test results for individual patients and is necessary when patients are followed radiographically over time. We present methods for testing both individual and population equivalence. The properties of the proposed methods are assessed in a Monte Carlo simulation study. Data from a mammography screening study is used to illustrate the proposed methods and compare them with results from more conventional methods of assessing equivalence and inter-procedure agreement.

Breast Neoplasms↗

Spiral CT scanning technique in the detection of aspiration of LEGO foreign bodies.

BACKGROUND: Radiolucent foreign bodies (FBs) such as plastic objects and toys remain difficult to identify on conventional radiographs of the neck and chest. Children may present with a variety of respiratory complaints, which may or may not be due to a FB. OBJECTIVE: To determine whether radiolucent FBs such as plastic LEGOs and peanuts can be seen in the tracheobronchial tree or esophagus using low-dose spiral CT, and, if visible, to determine the optimal CT imaging technique. MATERIALS AND METHODS: Multiple spiral sequences were performed while varying the CT parameters and the presence and location of FBs in either the trachea or the esophagus first on a neck phantom and then a cadaver. Sequences were rated by three radiologists blinded to the presence of a FB using a single scoring system. RESULTS: The LEGO was well visualized in the trachea by all three readers (both lung and soft-tissue windowing: combined sensitivity 89 %, combined specificity 89 %) and to a lesser extent in the esophagus (combined sensitivity 31 %, combined specificity 100 %). The peanut was not well visualized (combined sensitivity < 35 %). The optimal technique for visualizing the LEGO was 120 kV, 90 mA, 3-mm collimation, 0.75 s/revolution, and 2.0 pitch. This allowed for coverage of the cadaver tracheobronchial tree (approximately 11 cm) in about 18 s. Although statistical power was low for detecting significant differences, all three readers noted higher average confidence ratings with lung windowing among 18 LEGO-in-trachea scans. CONCLUSION: Rapid, low-dose spiral CT may be used to visualize LEGO FBs in the airway or esophagus. Peanuts were not well visualized.

Arachis↗

Multi-slice computed tomography as a screening tool for colon cancer, lung cancer and coronary artery disease.

Recent promising trials that use low-dose CT for the early detection of lung cancer have reinvigorated the interest in screening approaches. At the same time the development of fast image acquisition techniques, such as multislice CT, have sparked renewed interest in cardiac imaging within the radiological community. In addition to special cardiac capabilities, multislice CT has several other features such as high acquisition speed and low-dose requirements that may make this modality a universal radiological screening tool. Non-invasive disease detection is the radiologist's domain. In this paper we identify criteria for effective screening and apply these criteria to screening approaches with multislice CT when used for detection of three disease entities: colon cancer; lung cancer; and cardiovascular disease.

Colonic Neoplasms↗

Automatic segmentation of mammographic density.

RATIONALE AND OBJECTIVES: The purpose of this study was to evaluate a completely automatic method, based on Kittler's optimal threshold, to estimate breast density by using the mammographers' definition. MATERIALS AND METHODS: Thirty-two normal, right-craniocaudal-view mammograms of women aged 37-86 years were digitized. The whole breast area was segmented by using Kittler's optimal threshold procedure, and the dense portions were then segmented by using a modified version of Kittler's method. Segmentation results were validated by three independent mammographers who provided a signed percentage (in steps of 5%) to indicate the difference between their own visual estimation of the dense portions and the results obtained with the algorithm. The difference between the algorithm measurements and the mammographers' measurements was compared to the interobserver differences. RESULTS: A high correlation was found between the algorithm measured density and the mammographers' measurements. Spearman correlations ranged from 0.92 to 0.95 (P < .001). Algorithm-measured density differed from the mammographers' measurements by an average of 6.9% (ie, average of the absolute differences). In contrast, mammographers' measurements differed between themselves by an average of 5.4%. CONCLUSION: The difference between density as measured with the algorithm and as measured by the mammographers is similar to the differences observed between mammographers. This algorithm could be useful in providing clinically accurate estimates of breast density.

Adult↗

The "differential diagnosis" for multiple diseases: comparison with the binary-truth state experiment in two empirical studies.

RATIONALE AND OBJECTIVES: In practice readers must often choose between multiple diagnoses. For assessing reader accuracy in these settings. Obuchowski et al have proposed the "differential diagnosis" method, which derives all pairwise estimates of accuracy for the various diagnoses, along with summary measures of accuracy. The current study assessed the correspondence between the differential diagnosis method and conventional binary-truth state experiments. MATERIALS AND METHODS: Two empirical studies were conducted at two institutions with different readers and diagnostic tests. Readers used the differential diagnosis format to interpret a set of cases. In subsequent readings they interpreted the cases in binary-truth state experiments. Spearman rank correlation coefficients and the percentages of agreement in scores were computed, and the areas under the receiver operating characteristic curves were estimated and compared. RESULTS: The between-format Spearman rank correlation coefficients were 0.697-0.718 and 0.750-0.780 for the two studies; the between-reader correlations were 0.417 and 0.792, respectively. The percentages of agreement between formats for the two studies were 50.0%-51.7% and 72.9%-78.8%; the percentages of agreement between readers were 45.0% and 80%, respectively. In the first study there were several significant differences in the areas under receiver operating characteristic curves; in the second study these differences were small. CONCLUSION: The differences observed between the two formats can be attributed to within-reader variability and inherent differences in the questions posed to readers in the multiple-diagnoses versus binary-truth state reading sessions. The differential diagnosis format is useful for estimating accuracy when there are multiple possible diagnoses.

Confidence Intervals↗

Clinical evaluation of wavelet-compressed digitized screen-film mammography.

RATIONALE AND OBJECTIVES: The authors compared diagnostic accuracy and callback rates with conventional screen-film mammograms and wavelet-compressed digitized images. MATERIALS AND METHODS: Sixty sets of mammograms (four views per case) were digitized at a spatial resolution of 100 microm. The images were wavelet compressed to a mean compression ratio of 8:1 and reviewed by three mammographers. Five regions were evaluated in each breast. Suspicion of malignancy was graded on a scale of 0% to 100%, and receiver operating characteristic (ROC) analysis was performed. Callback rates were calculated by using the American College of Radiology's Breast Imaging Reporting and Data System lexicon scale. RESULTS: The mean diagnostic accuracy with compressed and conventional images was 0.832 and 0.860, respectively. The upper 95% confidence bound for the difference in ROC areas was 0.061. The mean false-positive rate at a fixed sensitivity of 0.90 was 0.041 for compressed images and 0.059 for conventional images. The mean callback rates for normal, benign, and malignant regions were 0.023, 0.305, and 0.677, respectively, for compressed images and 0.036, 0.447, and 0.750, respectively, for conventional images. The upper 95% confidence bound for the (absolute) differences in callback rates was 0.012 for normal regions, 0.163 for benign regions, and 0.138 for malignant regions. CONCLUSION: Diagnostic accuracies were equivalent for both compressed and conventional images. The mean false-positive rate at fixed sensitivity was much better with the compressed images. However, the callback rates for malignant lesions were lower when the compressed images were used.

Breast Diseases↗

Data analysis for detection and localization of multiple abnormalities with application to mammography.

RATIONALE AND OBJECTIVES: In assessing diagnostic accuracy it is often essential to determine the reader's ability both to detect and to correctly locate multiple abnormalities per patient. The authors developed a new approach for the detection and localization of multiple abnormalities and compared it with other approaches. MATERIALS AND METHODS: The new approach involves partitioning the image into multiple regions of interest (ROIs). The reader assigns a confidence score to each ROI. Statistical methods for clustered data are used to assess and compare reader accuracy. The authors applied this new method to a reader-performance study of conventional film images and digitized images used to detect and locate malignant breast cancer lesions. RESULTS: The ROI-based approach, the free-response receiver operating characteristic (FROC) curve, and the patient-based approach handle the estimation of the false-positive rate (FPR) quite differently. These differences affect the measures of the respective areas under the curves. In the ROI-based approach the denominator is the number of ROIs without a malignant lesion. In the FROC approach the average number of false-positive findings per patient is plotted on the x axis of the curve. In contrast, the patient-based approach mishandles the FPR by ignoring multiple detection and/or localization errors in the same patient. The FROC approach does not lend itself easily to statistical evaluations. CONCLUSION: The ROI-based approach appropriately captures both the detection and localization tasks. The interpretation of the ROI-based accuracy measures is simple and clinically relevant. There are statistical methods for estimating and comparing ROI-based estimates of accuracy.

Breast Neoplasms↗

Comparing the performance of mammographic enhancement algorithms: a preference study.

OBJECTIVE: The objective of this study was to compare the performance of four image enhancement algorithms on secondarily digitized (i.e., digitized from film) mammograms containing masses and microcalcifications of known pathology in a clinical soft-copy display setting. MATERIALS AND METHODS: Four different image processing algorithms (adaptive unsharp masking, contrast-limited adaptive histogram equalization, adaptive neighborhood contrast enhancement, and wavelet-based enhancement) were applied to one image of secondarily digitized mammograms of forty cases (10 each of benign and malignant masses and 10 each of benign and malignant microcalcifications). The four enhanced images and the one unenhanced image were displayed randomly across three high-resolution monitors. Four expert mammographers ranked the unenhanced and the four enhanced images from 1 (best) to 5 (worst). RESULTS: For microcalcifications, the adaptive neighborhood contrast enhancement algorithm was the most preferred in 49% of the interpretations, the wavelet-based enhancement in 28%, and the unenhanced image in 13%. For masses, the unenhanced image was the most preferred in 58% of cases, followed by the unsharp masking algorithm (28%). CONCLUSION: Appropriate image enhancement improves the visibility of microcalcifications. Among the different algorithms, the adaptive neighborhood contrast enhancement algorithm was preferred most often. For masses, no significant improvement was observed with any of these image processing approaches compared with the unenhanced image. Different image processing approaches may need to be used, depending on the type of lesion. This study has implications for the practice of digital mammography.

Aged↗

Sample size tables for receiver operating characteristic studies.

OBJECTIVE: I provide researchers with tables of sample size for multiobserver receiver operating characteristic (ROC) studies that compare the diagnostic accuracies of two imaging techniques. MATERIALS AND METHODS: I computed the number of patients and observers needed as a function of five parameters: the measure of diagnostic accuracy (area under the ROC curve, sensitivity at a false-positive rate </= 0.10, or specificity at a false-negative rate </= 0.10), conjectured level of accuracy, suspected difference in accuracy between the two imaging techniques, observer variability, and ratio of patients without to patients with the condition. RESULTS: The numbers of patients and observers required vary dramatically with these five parameters, increasing with more refined measures of accuracy, with lower accuracy levels, with smaller suspected differences, with greater observer variability, and with less balanced designs. The number of patients required for a study can be reduced by increasing the number of observers, and vice versa. When the intra- and interobserver variability is large, a study design with just four observers is usually inadequate. CONCLUSION: Many factors must be considered when determining the appropriate sample sizes for multiobserver ROC studies. My tables serve only as initial ballpark estimates. Investigators should compute sample size using parameters that reflect their clinical application.

ROC Curve↗

Quantitative analysis of breast parenchymal density: correlation with women's age.

RATIONALE AND OBJECTIVES: The authors evaluated the relationship between a woman's breast parenchymal density and her age by means of a quantitative method for measuring density from digitized mammograms. MATERIALS AND METHODS: The percentage of the breast considered to be dense was evaluated from mammograms of 50 women stratified by age. Quantitative analysis based on the computer segmentation of tissue in digitized mammograms was performed by three expert mammographers. The results of this analysis were compared with results from a review of the film mammograms by three expert mammographers. RESULTS: A slight decrease in the percentage of breast considered to be dense with increased age was observed. The average difference in the percentage of dense breast tissue between the youngest and the oldest age groups was 6.4% based on the digital review and 14.6% based on the film review. Within each age group, the total variability was on the order of 75%. CONCLUSION: The difference in mean magnitude between the youngest and oldest age groups was small and may not be clinically important. The variability within an age group was large, which suggests that age is not a reliable indicator of percentage of dense breast tissue.

Adult↗

Film-screen versus digitized mammography: assessment of clinical equivalence.

OBJECTIVE: The purpose of this study was to determine whether diagnostic accuracy and callback rates using digitized film images are equivalent to those using film-screen mammograms. MATERIALS AND METHODS: Sixty sets of mammograms (four views per case) were digitized at a spatial resolution of 100 microm. The images were reviewed by seven mammographers. Five regions were evaluated in each breast. Each region was scored on a scale of 0 100% for suspicion of malignancy, and a receiver operating characteristic analysis was performed. Callback rates were calculated using a published lexicon scale. RESULTS: The observers' mean diagnostic accuracies using films and digitized images were 0.872 and 0.848, respectively. The upper 95% confidence boundary on the difference in accuracy was 0.066. The mean callback rate for normal, benign, and malignant areas using films versus digitized images was 0.048 versus 0.055, 0.498 versus 0.441, and 0.786 versus 0.737, respectively. The upper 95% confidence boundary for the absolute difference in callback rates was 0.037, 0.026, and 0.130 for normal, benign, and malignant areas, respectively. CONCLUSION: The diagnostic accuracies of the digitized images and films were similar; however, an increase in callback rates of 0.037 (i.e., upper 95% confidence boundary) for normal results and a reduction in the callback rates of 0.130 for malignant lesions is important. The use of digitized film images, at a spatial resolution of 100 microm, may compromise patient treatment in clinical practice.

Breast Diseases↗

Correlation of valve closing to visualization of single-leg separation in radiographic images of Björk-Shiley convexo-concave heart valves.

BACKGROUND AND AIM OF THE STUDY: The study goal was to determine whether the visualization of single-leg separation (SLS) in cineangiographic sequences of Björk-Shiley convexo-concave heart valves could be correlated to the position of the occluder disk within the cardiac cycle. METHODS: Images from ten patient cases with SLS valves were reviewed by three experts, who identified the image frames within a cine sequence that appeared suspicious for SLS. The position of the occluder disk, the frame rate, and the length of the cardiac cycle were noted relative to these image frames. RESULTS: The probability of detecting a SLS was not significantly correlated to any of these factors. CONCLUSIONS: Visualization of SLS in cineangiographic images is limited to a few frames within an imaging sequence. It appears that other features within the image play a larger role in a clinician's ability to detect a fracture than do the cardiac dynamics of the system.

Cineangiography↗

On the comparison of correlated proportions for clustered data.

McNemar's test is often used to compare two proportions estimated from paired observations. We propose a method extending this to the case where the observations are sampled in clusters. The proposed method is simple to implement and makes no assumptions about the correlation structure. We conducted a Monte Carlo simulation study to compare the size and power of the proposed method with a test developed earlier by Eliasziw and Donner. In the presence of intracluster correlation, the size of McNemar's test can greatly exceed the nominal level. The size of Eliasziw and Donner's test is also inflated for some correlation patterns. The proposed method, on the other hand, is close to the nominal size for a variety of correlation patterns, although it is slightly less powerful than Eliasziw and Donner's procedure. The proposed method is a good alternative to Eliasziw and Donner's test when, in practice, little is known about the correlation pattern of the data.

Cluster Analysis↗

Postoperative portable chest radiographs: optimum use in thoracic surgery.

PURPOSE: Daily portable chest radiographs are routinely ordered in many institutions after thoracic surgery. Our purpose was to assess the efficacy and cost of this practice and to determine the optimum use of postoperative x-ray studies. METHODS: A prospective review of all portable chest x-ray films after 100 consecutive elective thoracotomies (DRG 75) was conducted. Each x-ray study initiated a three-part survey. First, the surgeon listed whether the x-ray study was routine and the anticipated management had it not been available. The radiologist then interpreted and scored the x-ray study as follows: negative, expected findings; A, minor findings necessitating no intervention; B, minor findings necessitating intervention; or C, major findings necessitating intervention. Finally, the x-ray film and the interpretation were returned to the surgeon. Any interventions necessitated by the x-ray study were recorded. RESULTS: In 6 months, 99 patients underwent 82 pulmonary resections and 18 other major procedures. In the postoperative period, 769 portable chest x-ray studies were ordered, median five per patient (range 2 to 49). Of these, 731 (95%) were routine and 38 (5%), nonroutine. Severity scores were as follows: negative in 106 (13.8%), A in 558 (72.5%), B in 59 (7.7%), and C in 46 (6.0%). X-ray findings altered management in 43 of 769 studies (5.6%): in 33 routine (4.5%), in 10 nonroutine (26.3%), in 13 A (2.3%), in 22 B (37.3%), and in 8 C (17.4%). CONCLUSIONS: These results demonstrate that routine daily portable chest x-ray studies have a minimal impact on management. It is, in fact, nonroutine x-ray studies that more often alter management. Had routine portable chest x-ray studies, which cost $114 each in our institution, been limited to one immediately after the operation, only 133 such studies (100 routine and 33 nonroutine) would have been needed in the care of these patients. Elimination of 636 (82.7%) x-ray studies reduces the cost of care by $725 per patient ($286,000 annually). For major thoracic procedures, it is safe, efficacious, and cost effective to eliminate routine postoperative portable chest x-ray studies and order nonroutine portable studies only when clinically indicated.

Adult↗

Confidence intervals for the receiver operating characteristic area in studies with small samples.

RATIONALE AND OBJECTIVES: The authors performed this study to address two practical questions. First, how large does the sample size need to be for confidence intervals (CIs) based on the usual asymptotic methods to be appropriate? Second, when the sample size is smaller than this threshold, what alternative method of CI construction should be used? MATERIALS AND METHODS: The authors performed a Monte Carlo simulation study where 95% CIs were constructed for the receiver operating characteristic (ROC) area and for the difference between two ROC areas for rating and continuous test results--for ROC areas of moderate and high accuracy--by using both parametric and nonparametric estimation methods. Alternative methods evaluated included several bootstrap CIs and CIs with the Student t distribution. RESULTS: For the difference between two ROC areas, CIs based on the asymptotic theory provided adequate coverage even when the sample size was very small (20 patients). In contrast, for a single ROC area, the asymptotic methods do not provide adequate CI coverage for small samples; for ROC areas of high accuracy, the sample size must be large (more than 200 patients) for the asymptotic methods to be applicable. The recommended alternative (bootstrap percentile, bootstrap t, or bootstrap bias-corrected accelerated method) depends on the estimation approach, format of the test results, and ROC area. CONCLUSION: Currently, there is not a single best alternative for constructing CIs for a single ROC area for small samples.

Confidence Intervals↗