A fully automated adaptive unsharp masking technique in digital chest radiograph.

The authors are developing a fully automated adaptive unsharp masking technique with parameters that depend on regional image features of a digital chest radiograph. A chest radiograph includes various regions such as lung fields, retrocardiac area, and spine. These areas have very different texture patterns and optical densities. Therefore, for best evaluation, it is necessary to enhance the image contrast of each region by an optimal parameter. In the current study, a chest radiograph was automatically divided into three segments (lung field, retrocardiac area, and spine) by using a histogram analysis of pixel values. The lung fields and retrocardiac area were selectively enhanced with a small mask size and mild weighting factors that had been previously determined to be optimal. The spine was enhanced with a large mask size and adequate weighting factors. An observer performance test indicated that this technique provides excellent diagnostic accuracy for simulated nodules in chest radiographs.     

Evaluation of optical unsharp masking and contrast enhancement of low-scatter chest radiographs

Conventional chest radiography poses a challenging technical problem because of its requirement for simultaneous high-contrast display and wide-latitude recording across the entire image. We developed and evaluated a method of producing chest radiographs by using a tantalum air-interspace grid for highly efficient scatter rejection, wide-latitude X-ray film for recording the low-scatter image, and a LogEtronics printer for optical unsharp masking and contrast enhancement of the recorded image (TWL technique). TWL images can be readily obtained and have excellent contrast and detail across the entire image. In comparison with a conventional technique, the TWL technique provides about a 15% improvement in image contrast in well-penetrated areas and a threefold to tenfold improvement in poorly penetrated areas. A detection study using simulated lung nodules and a chest phantom showed about 10% overall improvement in nodule detection with the TWL technique (51% vs 42%), most of which was due to improvement in detection rates in poorly penetrated areas of the chest (62% vs 26%). In well-penetrated areas, there was a decrease in detection rates (52% vs 44%) using TWL images despite measured improvements in image contrast in these areas. Possibly this was due to the observers' unfamiliarity with the reversed-contrast TWL images. Our results show the TWL technique to be valuable for improving image quality and diagnostic accuracy in chest radiography.     

Development of an automated patient-recognition method for digital chest radiographs using edge-enhanced images

It is important that all images in a picture archiving and communication system (PACS) environment should be stored in the correct location, e.g., in the proper patient's folder. However, if patient information, such as identification number or patient name, has been entered incorrectly, the image may be stored in the wrong place. We are developing an automated patient recognition method for chest radiographs based on a template-matching technique to prevent such filing errors. To further improve the performance of our method, we investigated the usefulness of a new automated patient-recognition method based on a template-matching technique by using edge-enhanced and smoothed images. We found that the relationship between the correlation values obtained with and without the edge-enhancement technique tended to provide different criteria for identifying correct or incorrect patients. When we combined the two methods to distinguish the images by a rule-based method, 67.1% of wrongly identified patients in our database could be identified as wrongly identified, without any false warnings for correctly identified patients. We consider that this automated method for patient recognition based on edge-enhanced images would be useful in preventing "wrong" images from being stored in a PACS environment.     

Evaluation of cardiopulmonary devices on chest radiographs: digital vs analog radiographs

We compared digital and analog radiographs of the chest for use in detecting and evaluating a variety of cardiopulmonary devices in 40 patients. The devices included 23 endotracheal tubes, 21 Swan-Ganz catheters, 14 central venous pressure catheters, 11 prosthetic valves, 10 chest tubes, six pacemaker wires, and five intraaortic balloon pumps. Each patient had at least one device, with a maximum of five devices (average, 2.3). Forty digital/analog film pairs were compared by five radiologists, who assigned confidence levels for various judgments about each device. The results showed that there were no statistically significant differences in the identification of the devices except for prosthetic valves (all valves were detected on digital radiographs, compared with 62% on analog radiographs). The devices were detected on 96% of the digital radiographs and 90% of the analog radiographs. Although digital and analog radiographs were comparable for detection of most devices, the digital radiographs allowed greater confidence in the identification of the devices and in the identification of the tip and course of Swan-Ganz and central venous pressure catheters.     

计算机辅助检测系统在数字化胸片肺结节检出中的应用价值

目的评价计算机辅助检测（computer-aided detection，CAD）系统在数字化胸片肺结节检出中的应用价值及其对放射科医师的辅助作用。方法选取数字化胸片328例。由2名专家组医师应用IQQATM-Chest系统阅读具有结节样阴影的胸片，2人意见达成一致后标记结节的位置和大小并保存标记结果，将标记结果作为金标准来评估CAD系统检测肺结节的能力。由8名不同年资的放射科医师首先独立阅读具有结节样阴影的数字X线摄影（DR）胸片并保存诊断结果，然后再应用CAD系统阅读胸片，将最终结果存入CAD系统。应用受试者操作特征曲线（ROC）和配对t检验来分析放射科医师应用CAD系统前后在肺结节检测能力上的差异。结果在100例DR胸片中，金标准结节151个，CAD系统肺结节检测敏感性为78．1％（118．0／151），低年资放射科医师不用和应用CAD系统时，肺结节的检测敏感性分别为62．4％（94．2／151）和77．4％（116．8／151），ROC曲线下面积分别为0．769和0．836，二者之间的差异具有统计学意义（P〈0．01）；高年资放射科医师不用和应用CAD系统时，肺结节的检测敏感性分别为73．8％（111．5／151）和76．2％（115．0／151），ROC曲线下面积分别为0．820和0．827，二者之间的差异无统计学意义（P〉0．05）。结论CAD系统能够辅助放射科医师提高肺小结节的检测敏感性，对低年资医师的帮助更大。     

Computer-simulated lung nodules in digital chest radiographs for detection studies

Computer simulations of lung nodules overcome many shortcomings of creating radiographs using anthropomorphic nodule phantoms for lung nodule detection studies, but these algorithms can be cumbersome and involved. A simple, fast, and flexible computer program to simulate lung nodules in digital chest radiographs for detection studies is reported. To verify the realism of the simulated nodules, a psychophysical study and a statistical study were conducted. In the psychophysical study, six radiologists and four nonradiologists were asked to distinguish between 17 real lung nodules and 17 computer-simulated lung nodules shown in eight radiographs. The results show that the computer-simulated lung nodules are indistinguishable visually from real lung nodules. Using parameters from the Rose model of vision, results show that the simulated and real nodules are the same statistically. Thus, besides visual validity, statistical analysis in confirming the validity of the simulated lung nodules is included.     

Evaluation of usefulness of color digital summation radiography for solitary pulmonary nodules on chest radiographs

Purpose The aim of this study was to evaluate the usefulness of novel color digital summation radiography (CDSR) for detecting solitary pulmonary nodules on chest radiographs by observers with different levels of experience. Materials and methods A total of 30 healthy controls and 30 patients with newly detected solitary pulmonary nodules were evaluated. Six radiologists and five residents evaluated three image sets: set A, current and prior radiographs only; set B, set A with temporal subtraction images; and set C, set A with CDSR. The observers were asked to rate each image set using a continuous rating scale. In addition, the reading time required for each set was recorded. Results The radiologists showed no significant differences in the mean A_z value between set A, set B, and set C. However, the residents showed significant differences between set A and set B and between set A and set C. In addition, for set B and set C, the mean reading time per case of all readers was significantly shorter than that for set A. Conclusion The detection capability of observers with little experience is comparable to that of experienced observers when reading radiographs with temporal subtraction images or with CDSR. The usefulness of CDSR is comparable to that of temporal subtraction.     

Quantitative analysis of pneumoconiosis in standard chest radiographs]

A computerized method to quantify and characterize interstitial diseases by using physical texture measures obtained from an analysis of the power spectrum of lung textures in digital chest radiographs was applied to Japanese standard radiographs of pneumoconiosis. Texture measures were determined from standard radiographs of silicosis, asbestosis, and other types of pneumoconiosis as well as chest radiographs of normal lungs. Our preliminary results indicated that the texture measures obtained from computer analysis corresponded closely with the standard categories of silicosis. However, there was no significant correlation between texture measures and the categories for asbestosis and other types of pneumoconiosis in terms of texture pattern. Japanese standard radiographs of pneumoconiosis are categorized according to the profusion of opacities, without reference to the size and shape of the opacities. Furthermore, in some films the size and shape of the opacities vary considerably within the same category. Therefore, it is considered that these characteristics of the standard films affected the results of our texture measures. It also considered that a large ROI and other texture measures are needed to characterize large opacities and mixed-shaped opacities of pneumoconiosis.     

Evaluation of competence in the interpretation of chest radiographs

RATIONALE AND OBJECTIVES: The purpose of this study was to determine relative rates of missed diagnoses for radiologists as a measure of competence in interpreting chest radiographs. MATERIALS AND METHODS: Cases involving differing interpretations of chest radiographs were collected from January 1994 through December 1999 by faculty (chest and nonchest radiology specialists) in an academic radiology department. A quarterly peer-review process designated cases months after the fact, and anonymously, as no miss or as class I (nondiagnosable), class II (very difficult diagnosis), class III (should be diagnosed most of time), or class IV (should almost always be diagnosed) missed diagnoses. The rates and classes of missed diagnoses were compared among chest faculty and for the nonchest radiology specialists as a group. RESULTS: Chest radiologists read 184,977 studies, and nonchest radiologists read 300,684 studies. Of these, 243 missed diagnoses were classified (classes I and II, 184 cases; class III, 50; and class IV, nine). No difference was detected in the rate of class III and IV misses among chest faculty, but nonchest faculty had significantly more class III (P = .022) and class IV misses (P = .016). CONCLUSION: Random sampling of differing interpretations can yield a relative rate of missed diagnoses for radiologists. No difference was detected in clinically important misses (ie, classes III and IV) among chest radiologists, but a statistically significantly higher rate of seemingly obvious misdiagnoses was found for nonchest specialty radiologists. Potential biases may have influenced this analysis, including disease prevalence, sampling, clinical factors, observer variability, and truth-in-diagnosis.     

Detection of lung nodules on digital chest radiographs: potential usefulness of a new contralateral subtraction technique

PURPOSE: To evaluate the potential usefulness of a contralateral subtraction technique developed for radiologists' performance in the detection of subtle lung nodules on chest radiographs. MATERIALS AND METHODS: Fifty chest radiographs (25 normal and 25 abnormal with a subtle lung nodule) that were digitized with a 0.175-mm pixel size and 4,096 gray levels were used. Twelve radiologists (10 attending and two residents) participated in observer tests and read both original and contralateral subtraction images with a sequential testing method. Radiologists' performance was evaluated by means of receiver operating characteristic analysis with use of a continuous rating scale. The beneficial and detrimental effects of the contralateral subtraction technique on the radiologists' performance were also evaluated. RESULTS: The area under the receiver operating characteristic curve values obtained without and with contralateral subtraction images were 0.926 and 0.962, respectively. Results indicated that the contralateral subtraction images significantly (P <.05) improved diagnostic accuracy, particularly for radiologists with limited experience. CONCLUSION: The contralateral subtraction technique can assist radiologists in the correct identification of subtle lung nodules on chest radiographs.     

Spatial resolution requirements for digital chest radiographs: an ROC study of observer performance in selected cases

Thirty-eight selected clinical radiographs were digitized and displayed on a 1,024-line monitor at pixel sizes of 1.6, 0.8, 0.4, and 0.2 mm. Eighteen experienced radiologists assessed the radiographs and digital images, which included 12 examples of abnormal solitary nodular density, ten examples of septal lines, and 16 controls, six of which showed diffuse lung abnormalities. For each level of spatial resolution and for film reading, observers gave their decision confidence on a sliding scale of probability. Receiver operating characteristic curves were generated from these data. It was found that while spatial resolution requirements for solitary nodules were not critical for pixel sizes at or below 0.8 mm, the requirement for septal lines was likely to be 0.4 mm (1.25 line pairs/mm).     

Potential usefulness of biological fingerprints in chest radiographs for automated patient recognition and identification

RATIONALE AND OBJECTIVES: The purpose of this study was to demonstrate the potential usefulness of "biological fingerprints" in chest radiographs for automated patient recognition and identification. MATERIALS AND METHODS: Thoracic fields, cardiac shadows, the superior mediastinum, lung apices, a part of the right lung, and the right lower lung that includes the costophrenic angle were used as biological fingerprints in chest radiographs. Each of the biological fingerprints in a current chest radiograph was used as a template for determination of the correlation value with the corresponding biological fingerprint in a previous chest radiograph for patient recognition and identification. The overall performance of the method developed was examined in terms of receiver operating characteristic curves. RESULTS: Receiver operating characteristic curves obtained with different biological fingerprints, except for the part of the right lung, indicated a high performance in identifying patients. These results showed that a new concept of biological fingerprints in radiologic images would be useful in patient recognition and identification. The low performance with the part of the right lung seems to be related to a general observation that this region does not usually include features unique to a specific patient. The performance of the artificial neural networks by use of a combination of five biological fingerprints was higher than results obtained with each biological fingerprint. CONCLUSION: The use of automated patient identification based on biological fingerprints in chest radiographs is promising for helping to discover misfiled patient images, especially in a picture archiving and communication system environment.     

ROC analysis of detection of interval changes in interstitial lung diseases on digital chest radiographs using the temporal subtraction technique

PURPOSE: To demonstrate the clinical usefulness of a temporal subtraction technique for the detection of interval changes in various interstitial lung diseases on digital chest radiographs. MATERIALS AND METHODS: One hundred pairs of chest radiographs in 34 patients (63 with and 37 without interval changes) with various interstitial lung diseases were selected. All cases were confirmed by serial chest computed tomography (CT) and ascertained by radiologists. All chest radiographs were obtained with a computed radiography (CR) system, and temporal subtraction images were produced with an iterative image-warping technique. Four radiologists and two thoracic physicians provided confidence levels for interval changes in interstitial lung diseases with and without temporal subtraction. Their performances with and without temporal subtraction were evaluated by means of receiver operating characteristic (ROC) analysis using a sequential test. RESULTS: The area under the ROC curve (Az) values of six observers obtained with and without temporal subtraction were 0.90 and 0.78, respectively. Results showed that the detection of interval changes in interstitial lung diseases was significantly improved by the use of temporal subtraction images compared with CR images alone(P = 0.002). Furthermore, the high detection rate was achieved with temporal subtraction images regardless of the subtlety and location of interval changes. CONCLUSION: Temporal subtraction improved the diagnostic accuracy of radiologists in detecting interval changes in interstitial lung diseases on chest radiographs. It was also useful for cases of multiple interval changes.     

Unsharp masking of low-dosed digital luminiscence radiographs: results of a receiver operating characteristics analysis

The extent to which the diagnostic performance of low-dosed storage phosphors can be improved by suitable image postprocessing was investigated. Storage phosphors radiographs and conventional 200- and 400-speed screen-film combinations of an antropomorphic chest phantom with simulated nodules and linear structures were compared. The detection of test bodies was determined in an ROC analysis based on 16 200 individual observations. The two analog screen-film systems did not differ significantly. As far as the storage phosphor radiographs are concerned, a higher diagnostic performance was achieved with a large filter kernel (S 40), whereas small filter kernels and edge-enhanced imaging modes reduced the detection of detail. The image quality of low-dosed storage phosphors is thus similar to high-amplification screen-film combinations if large filter kernels are used in postprocessing of the image. Correspondence to: R.-D. Müller     

Development of computerized method for automated classification of the body parts in digital radiographs

PURPOSE: In picture archiving and communication systems (PACS), the information on the body parts included in radiographs is often not or incorrectly recorded in an image header. In order to apply the computer-aided diagnosis (CAD) system in the PACS environment, the body parts in radiographs need to be recognized correctly by computer. The purpose of this study is to develop a computerized method for correctly classifying the body parts in digital radiographs based on a template matching technique.METHODS/MATERIALS: The image database used in this study was 1032 digital radiographs (14 x 17 inches) obtained with a computed radiography, and included 505 chest of postetroanterior view, 39 chest of lateral view, 241 abdomen, 108 pelvis, 10 upper limbs, 125 lower limbs, and 4 thoracic spine. In this method, test images were classified into four body parts, i.e., (1) chest, (2) abdomen, (3) pelvis, and (4) upper/lower limbs and thoracic spine. This computerized method was tested with 852 images, since 180 images were employed for creation of 98 templates, which represented the average radiographs for various body parts. Our approach was to examine the similarity of a given test image with templates by use of the cross-correlation values as the similarity measures. The body part of the test image was identified as the body part in the template yielding the maximum correlation value. Our method consisted of the following five steps. First, test images were classified into one of three groups; i.e. 1) chest and abdomen, 2) pelvis, and 3) upper/lower limbs and thoracic spine by using the templates obtained from images with the average size and position. Second, the remaining uncertain images were classified by using additional templates in various directions. Third, the chest and abdomen group was separated into two subgroups; i.e.chest and abdomen. Fourth, in order to classify some uncertain images, templates were shifted horizontally and vertically. Fifth, outer pixels of templates were eliminated to avoid the misclassification due to x-ray collimation. RESULTS: Our preliminary results indicated that the body parts for 850 cases (99.8%) were correctly classified with our method. CONCLUSIONS: This method would be useful for automated identification of the body parts in radiographs when various CAD systems would be implemented in the PACS environment.