How well can radiologists using neural network software diagnose pulmonary embolism?

OBJECTIVE: This study evaluated and optimized the performance of an automated artificial neural network image interpreter in the diagnosis of pulmonary embolism on ventilation-perfusion lung scans. The computer interpretations were compared with the interpretations of three experienced observers. MATERIALS AND METHODS: Digital data were obtained from 100 patients with normal findings on chest radiographs who were undergoing both radionuclide ventilation-perfusion scanning and pulmonary angiography. Interpretations of differently trained neural networks were compared with those of three experienced nuclear medicine practitioners unaware of the clinical diagnosis. RESULTS: Machines running neural networks performed similarly to experienced scan interpreters in the detection of pulmonary embolism. Both the human observers and the networks performed best in cases with large emboli. Neural network performance was best in the right lung, when the networks were trained using only cases with large emboli and when networks were trained independently in the right and left lungs. The best predictions resulted from a collaborative interpretation incorporating both the human and computer predictions. CONCLUSION: Computers running artificial neural networks using scan data obtained directly from the anterior and posterior ventilation and perfusion images, without human involvement, perform comparably with experienced observers in patients with normal findings on chest radiographs. Human observers can improve their interpretations by incorporating computer output to formulate diagnostic prediction. The method of training the networks is critical to optimizing performance.     

The lung scan and the abnormal chest X-ray: difficult diagnoses

PURPOSE: The ventilation-perfusion lung scan is evolving into an alternative diagnostic test for patients who cannot tolerate contrast computed tomography studies. It is likely that this shift in patient population will result in a larger proportion of patients with abnormal chest radiographs. The purpose of this study was to determine whether a computer based scan analysis could assist clinical interpretation in this diagnostically difficult population. METHODS: Radionuclide ventilation-perfusion (V-P) images were obtained from 118 patients with normal chest radiographs and 144 patients with abnormal radiographs who underwent pulmonary angiography within 3 days of the radionuclide study. Artificial neural networks (ANNs) were created using only objective image-derived inputs to diagnose the presence of pulmonary embolism. The ANN predictions were compared with clinical scan interpretations and with the results of angiography. RESULTS: In both patients with normal and with abnormal chest radiographs, the ANN based method performed comparably to the clinical interpretation of record. An average of the clinical and ANN estimates of the likelihood of embolism was more accurate than was either method alone. CONCLUSIONS: Computer-based V-P scan analysis performs comparably to clinical interpretation for patients with abnormal chest radiographs. The different analytical perspective of the digital method improves test performance when used in conjunction with standard clinical interpretation. This can improve the chances of reaching a diagnosis for patients who lack diagnostic alternatives to the V-P scan.     

Predicting the presence of acute pulmonary embolism: a comparative analysis of the artificial neural network,logistic regression,and threshold models

OBJECTIVE: The objective of this study was to determine whether an artificial neural network, a new data analysis method, offers increased performance over conventional logistic regression in predicting the presence of a pulmonary embolism for patients in a well-known data set. MATERIALS AND METHODS: Data from the 1064 patients who received an angiographically based diagnosis of pulmonary embolism in the Prospective Investigation of Pulmonary Embolism Diagnosis study were encoded using a previously described method. The 21 input variables represented abnormalities identified on each patient's ventilation-perfusion scan and chest radiograph. Two methods-an artificial neural network with one hidden layer and a multivariate logistic regression-were compared for accuracy in predicting the presence or absence of pulmonary embolism on subsequent pulmonary arteriography. RESULTS: No significant difference was observed between the two methods. Areas under the receiver operating characteristic curves +/- standard deviation were 0.78 +/- 0.02 for the artificial neural network model and 0.79 +/- 0.02 for the logistic regression model. Furthermore, use of these two methods resulted in no more diagnostic accuracy than did the use of a simple threshold model based only on the number of subsegmental perfusion defects, which was the dominant input variable. CONCLUSION: In the study population, the usefulness of data from ventilation-perfusion scans as predictors of the presence of a pulmonary embolism was similar for the three analytic methods, a finding that reinforces the importance of making comparisons to simpler or more established methods when performing studies involving complex analytic models, such as artificial neural networks.     

Pulmonary embolism revealed on helical CT angiography: comparison with ventilation-perfusion radionuclide lung scanning

OBJECTIVE: We compared helical CT angiography and ventilation-perfusion radionuclide lung scanning as initial tests in the diagnosis of acute pulmonary embolism. SUBJECTS AND METHODS: Two hundred sixteen consecutive patients who were clinically suspected of having acute pulmonary embolism underwent helical CT angiography, ventilation-perfusion radionuclide lung scanning, and Doppler sonography of the veins of the legs. On the basis of concordance of the results for ventilation-perfusion radionuclide lung scanning and helical CT angiography and on the degree of clinical suspicion, certain patients underwent pulmonary angiography. Patients without pulmonary embolism at initial evaluation in whom no treatment was instituted were followed up for at least 3 months to determine the potential recurrence of thromboembolic disease. RESULTS: Of the 216 patients, 37 (17%) were excluded because of insufficient data to assess the initial event. Final diagnosis for the 179 remaining patients was pulmonary embolism in 68 (37.9%) and no pulmonary embolism in 111 (62.0%), based on pulmonary angiography in 23 patients (12.8%) and concordant imaging findings and outcome in the remaining patients. Statistically significant differences (p < 0.05) were found between sensitivity, specificity, positive predictive value, and negative predictive value for helical CT angiography and ventilation-perfusion radionuclide lung scanning (94.1% versus 80.8%; 93.6% versus 73.8%; 95.5% versus 82%; and 96.2% versus 75.9%, respectively). Interobserver agreement was excellent for helical CT angiography (kappa = 0.72) and moderate for ventilation-perfusion radionuclide lung scanning (kappa = 0.22). CONCLUSION: Helical CT angiography could replace ventilation-perfusion radionuclide lung scanning as the initial test for screening patients who are clinically suspected of having pulmonary embolism.     

Pulmonary embolism of polymethyl methacrylate during percutaneous vertebroplasty and kyphoplasty

OBJECTIVE: The purpose of this study was to determine the frequency, radiographic findings, and clinical significance of a pulmonary embolism of cement occurring during percutaneous vertebroplasty or kyphoplasty as detected on conventional chest radiography. MATERIALS AND METHODS: Chest radiographs were obtained after 69 percutaneous vertebroplasty procedures in 64 patients. Chest radiographs were reviewed retrospectively for the presence of pulmonary emboli of cement, and findings were assessed. The frequency was calculated from the cases treated. Medical records were reviewed for procedure-related complications. RESULTS: The emboli of cement were noted radiographically in three (4.6%) of 65 procedures performed in our institution. All patients with cement emboli had multiple myeloma. The chest radiographic findings were multiple radiographically dense opacities with a tubular and branching shape that were scattered sporadically or distributed diffusely throughout the lungs. All patients with cement pulmonary embolism remained asymptomatic. A correlation of embolism of cement to lungs was found with paravertebral venous cement leak (p < 0.001) but not with the number of vertebral bodies treated (p = 0.185) or with the type of procedure performed-kyphoplasty versus vertebroplasty (p = 0.98). CONCLUSION: Pulmonary embolism of cement is seen in 4.6% of patients after percutaneous vertebroplasty or kyphoplasty. The characteristic radiographic findings should be recognized by radiologists.     

Ventilation imaging with Kr-81m

Sixty-five patients with suspected pulmonary embolism were studied prospectively with both Kr-81 m and Xe-133 ventilation imaging and Tc-99m MAA perfusion imaging. The krypton images, perfusion scintigrams and chest radiographs were read independently of the xenon images, perfusion scintigrams and chest radiographs by three observers. The studies of 53 patients were interpreted as normal or as indicative of a low or intermediate probability for pulmonary embolism with both gases. One study indicated intermediate probability with Xe-133 due to diffuse, severe xenon retention but low probability with Kr-81 m because of close ventilation-perfusion correspondence. The studies of 9 patients indicated a high probability of embolism with both gases, while those of two additional patients (one with emboli at angiography) indicated a high probability only with Kr-81m. While essential agreement between Xe-133 and Kr-81m ventilation imaging was found in most patients, the significant difference in interpretation in 2 of 11 patients with probable pulmonary embolism suggests that a controlled, prospective trial with pulmonary angiography is warranted before Kr-81m is employed for routine clinical use.     

Pulmonary perfusion patterns and pulmonary arterial pressure

PURPOSE: To use artificial intelligence methods to determine whether quantitative parameters describing the perfusion image can be synthesized to make a reasonable estimate of the pulmonary arterial (PA) pressure measured at angiography. MATERIALS AND METHODS: Radionuclide perfusion images were obtained in 120 patients with normal chest radiographs who also underwent angiographic PA pressure measurement within 3 days of the radionuclide study. An artificial neural network (ANN) was constructed from several image parameters describing statistical and boundary characteristics of the perfusion images. With use of a leave-one-out cross-validation technique, this method was used to predict the PA systolic pressure in cases on which the ANN had not been trained. A Pearson correlation coefficient was determined between the predicted and measured PA systolic pressures. RESULTS: ANN predictions correlated with measured pulmonary systolic pressures (r = 0.846, P <.001). The accuracy of the predictions was not influenced by the presence of pulmonary embolism. None of the 51 patients with predicted PA pressures of less than 29 mm Hg had pulmonary hypertension at angiography. All 13 patients with predicted PA pressures greater than 48 mm Hg had pulmonary hypertension at angiography. CONCLUSION: Meaningful information regarding PA pressure can be derived from noninvasive radionuclide perfusion scanning. The use of image analysis in concert with artificial intelligence methods helps to reveal physiologic information not readily apparent at visual image inspection.     

Comparison of Biello, McNeil, and PIOPED criteria for the diagnosis of pulmonary emboli on lung scans

The McNeil, Biello, and newly proposed PIOPED (from the National Institutes of Health-sponsored study, Prospective Investigation of Pulmonary Embolism Detection) interpretive methods for detection of pulmonary embolism on lung scans were compared in 96 patients who also underwent pulmonary angiography. Segmental findings on 99mTc perfusion and aerosol ventilation scans, chest radiographs, and pulmonary angiograms obtained within 48 hr of each other were encoded along with other information into a data base to facilitate analysis. The McNeil, Biello, and PIOPED criteria were applied to the encoded data. Although the PIOPED set of criteria yielded the most favorable likelihood ratio for predicting an angiogram showing pulmonary emboli and a favorable likelihood ratio for predicting an angiogram not showing pulmonary emboli, it had the highest number of indeterminate studies. The McNeil criteria demonstrated the least favorable likelihood for predicting pulmonary emboli on an angiogram. The Biello and McNeil criteria showed the most favorable likelihood ratio for predicting an angiogram not showing pulmonary emboli. Analysis of receiver-operating-characteristic (ROC) curves yielded the greatest area under the ROC curve for the Biello criteria, but there were no statistically significant differences among the three sets of criteria. This study suggests that the Biello scheme represents the best compromise of the sets of criteria studied.     

Pulmonary cement embolism: a complication of percutaneous vertebroplasty

Background: Percutaneous vertebroplasty (PV) has recently become a very common procedure for vertebral compression fractures. Extravasation of cement, a common event associated with vertebroplasty, may lead to cement emboli in the lungs.

Purpose: To determine the frequency of pulmonary cement embolism after percutaneous vertebroplasty.

Material and Methods: Between 2002 and 2006, 128 percutaneous vertebroplasties were performed in 73 patients (56 women and 17 men) in our institution. Postprocedural chest radiographs were obtained for all patients and assessed for the presence of pulmonary cement emboli.

Results: Pulmonary cement embolism was detected on chest radiographs and confirmed with chest computed tomography (CT) in four patients treated with percutaneous vertebroplasty for osteoporotic collapse and one patient treated for multiple myeloma. The imaging finding of pulmonary cement embolism was solitary or multiple fine radiodense lines with occasional branching patterns. The frequency of pulmonary cement embolism was 6.8%.

Conclusion: An incidence of pulmonary cement embolism of 6.8% during PV was found. Close clinical follow-up, postprocedural chest radiographs, and chest CT scans, if necessary, are important for the detection of pulmonary cement embolism at an early stage.     

Clinical validity of helical CT being interpreted as negative for pulmonary embolism: implications for patient treatment.

OBJECTIVE: The purpose of our study was to assess the clinical usefulness of helical CT findings that are interpreted as negative for pulmonary embolism. MATERIALS AND METHODS: One hundred twenty-six patients underwent 132 helical CT examinations and 352 patients underwent ventilation-perfusion scanning for suspected acute pulmonary embolism over a 17-month period at a single institution. Findings from clinical follow-up at a minimum of 6 months were assessed, with a special focus on the presence of recurrent thromboembolism and mortality in 78 consecutive patients in whom helical CT findings were interpreted as negative for pulmonary embolism and anticoagulant therapy was not administered (group I). During the same 17-month period, 46 patients underwent ventilation-perfusion scanning that was interpreted as normal (group II), and 132 patients underwent ventilation-perfusion scanning that was interpreted as showing a very low to low probability for pulmonary embolism (group III). Patients in groups II and III did not undergo helical CT or pulmonary angiography and did not receive anticoagulant therapy. However, clinical follow-up was solicited. Patients from groups II and III were used as control subjects. RESULTS: Nine patients in group I died, one of whom was found to have a microscopic pulmonary embolism at autopsy. In group II, four patients died, none of whom were shown to have a missed or recurrent pulmonary embolism. Of the 18 patients in group III who died, three had a recurrent or missed pulmonary embolism (mean interval, 9 days), and two were found to have deep vein thrombosis on sonography of the leg (mean interval, 12 weeks). Negative predictive values for helical CT, normal lung scanning, and low-probability ventilation-perfusion scanning were 99%, 100%, and 96%, respectively (p = .299). CT provided either additional findings or an alternate diagnosis in 42 (53.8%) of the 78 patients in whom helical CT findings had been interpreted as negative for pulmonary embolism. CONCLUSION: A helical CT scan can be effectively used to rule out clinically significant pulmonary emboli and may prevent further investigation or unnecessary treatment of most patients.     

Pulmonary embolism with unilateral lung scan defects and matching infiltrates.

Ten patients with matching ventilation-perfusion lung scan defects and corresponding pulmonary infiltrates were evaluated with segmental pulmonary angiography. All ten patients presented with sudden onset of pleuritic chest pain and fever. Pulmonary emboli were documented in three of the ten patients (30%). The remaining seven patients had pneumonia or atelectasis. The findings emphasize the non-diagnostic nature of lung scans which show only matching ventilation and perfusion defects in regions of pulmonary infiltrates. Segmental pulmonary angiography is recommended for differentiating pulmonary embolism from atelectasis or pneumonia in these patients.     

Role of chest radiography in predicting the extent of airway disease in patients with suspected pulmonary embolism

An independent evaluation of the chest radiographs and ventilation-perfusion (VP) scintigrams of 119 patients (121 studies) was performed to assess the ability of radiographic findings to indicate the extent of ventilation abnormalities in patients suspected of having pulmonary embolism (PE). VP scans were obtained with krypton-81m in 72 cases and with xenon-133 in 49. All scintigrams showing ventilation abnormalities in greater than 50% of the lung were considered indeterminate for PE, owing to an increased likelihood of false-negative interpretations, as reported in previous studies. Indeterminate VP studies were found in all 21 patients who had radiographic findings of widespread obstructive pulmonary disease (OPD), in 35% with radiographic findings of focal OPD (P less than .001), and in only 18% with no OPD seen radiographically (P less than .001). VP scintigraphy is likely to be indeterminate for PE when widespread OPD is seen by chest radiography and ventilation imaging may not be warranted in these patients.     

N-butyl-2-cyanoacrylate pulmonary embolism after endoscopic injection sclerotherapy for gastric variceal bleeding

PURPOSE: The purpose of this work was to describe the radiologic and clinical manifestations of n-butyl-2-cyanoacrylate pulmonary embolism (PE) after endoscopic injection sclerotherapy (EIS) for gastric variceal bleeding. METHOD: From 1992 to 1999, the medical records of 140 patients who had undergone EIS using n-butyl-2-cyanoacrylate were reviewed for identification of respiratory symptoms and amount of injection, and their pre- and postprocedure chest radiographs were reviewed to identify PE. In patients with PE, pre- and postprocedure chest radiographs (6/6), chest CT scans (3/6), lung perfusion scans (3/6), and follow-up chest radiographs (6/6) were analyzed retrospectively. RESULTS: Radiographically evident PE was observed in 6 (4.3%) of 140 patients. In comparison with patients without emboli, these patients received a higher mean volume of injection (4.2 vs. 1.8 ml) (p = 0.0011). Four of the six patients with pulmonary emboli had respiratory symptoms. Chest radiographs and CT scans showed unusual tubular or nodular, radiopaque pulmonary emboli along the pulmonary vessels. Multiple peripheral, wedge-shaped, subsegmental perfusion defects were seen on perfusion lung scans. In five of six patients, the radiographic abnormalities showed complete or partial resolution. There were no fatalities directly associated with PE. CONCLUSION: Radiographically evident PEs are uncommonly observed following EIS and appear to be more common in patients receiving a higher volume of liquid acrylate. Affected patients were either mildly symptomatic or asymptomatic, and there were no direct fatalities of this complication.     

Symmetrical perfusion defects without pulmonary embolism

We report two patients with ventilation-perfusion (V-P) images that demonstrate large symmetric perfusion defects with normal chest radiographs and ventilation images who did not have pulmonary embolism (PE) by angiography. This experience suggests that patients with symmetrical V-P mismatches should not be assumed to have PE. The need for pulmonary angiography in such patients should be carefully considered.This work was supported in part by National Institutes of Health SCOR in Thrombosis HL 14147     

Pulmonary lymphangioleiomyomatosis: correlation of ventilation-perfusion scintigraphy, chest radiography, and CT with pulmonary function tests

PURPOSE: To determine the findings on ventilation-perfusion (V-P) scintigrams, computed tomographic (CT) scans, and chest radiographs and correlate them with pulmonary function test results in patients with lymphangioleiomyomatosis. MATERIALS AND METHODS: V-P scintigraphy, chest radiography, conventional and thin-section CT, and pulmonary function tests were performed in 39 patients. The images were graded on a scale of 0 (normal) to 3 (severely abnormal). RESULTS: Imaging abnormalities were found on 92% of ventilation scintigrams, 92% of perfusion scintigrams, 79% of chest radiographs, 100% of CT scans, and 100% of thin-section CT scans. On ventilation scintigrams, 28 (72%) patients demonstrated a speckling pattern. On CT scans, all patients had pulmonary cysts. Univariate analysis showed that extent of disease on chest radiographs and CT scans, cyst size, V-P abnormalities, and degree of speckling were inversely correlated with forced expiratory volume in one second (FEV(1)), diffusing capacity of lung for carbon monoxide, and the ratio of FEV(1) to forced vital capacity (FVC) (P <.01) but not with FVC and total lung capacity. Larger cyst size correlated with extent of disease at CT, but not significantly (P =.056). CONCLUSION: Scintigraphic and radiologic abnormalities are seen in a majority of patients with lymphangioleiomyomatosis. On ventilation scintigrams, a frequently seen speckling pattern may be related to accumulation of radionuclide in pulmonary cysts-a hallmark of the disease at CT. Findings with each imaging modality correlate with certain pulmonary functions.     

Effect of an artificial neural network on radiologists' performance in the differential diagnosis of interstitial lung disease using chest radiographs

OBJECTIVE: We developed a new method to distinguish between various interstitial lung diseases that uses an artificial neural network. This network is based on features extracted from chest radiographs and clinical parameters. The aim of our study was to evaluate the effect of the output from the artificial neural network on radiologists' diagnostic accuracy. MATERIALS AND METHODS: The artificial neural network was designed to differentiate among 11 interstitial lung diseases using 10 clinical parameters and 16 radiologic findings. Thirty-three clinical cases (three cases for each lung disease) were selected. In the observer test, chest radiographs were viewed by eight radiologists (four attending physicians and four residents) with and without network output, which indicated the likelihood of each of the 11 possible diagnoses in each case. The radiologists' performance in distinguishing among the 11 interstitial lung diseases was evaluated by receiver operating characteristic (ROC) analysis with a continuous rating scale. RESULTS: When chest radiographs were viewed in conjunction with network output, a statistically significant improvement in diagnostic accuracy was achieved (p < .0001). The average area under the ROC curve was .826 without network output and .911 with network output. CONCLUSION: An artificial neural network can provide a useful "second opinion" to assist radiologists in the differential diagnosis of interstitial lung disease using chest radiographs.     

Multiple coarctations of the pulmonary artery: scintigraphic appearance

A case of angiography-proven multiple coarctations of the pulmonary arteries is presented. The patient had a history of long-standing exertional dyspnea and intranasal cocaine use. A chest radiograph suggested oligemia in the left lung and enlargement of the right pulmonary artery, prompting ventilation and perfusion radionuclide lung imaging. The combined scintigraphic and radiographic findings were indistinguishable from those caused by pulmonary emboli. Although rare, this entity should be included in the differential diagnosis of multiple unmatched perfusion defects on pulmonary ventilation-perfusion studies.     

Idiopathic pulmonary fibrosis. A rare cause of scintigraphic ventilation-perfusion mismatch

A case of idiopathic pulmonary fibrosis with multiple areas of mismatch on ventilation-perfusion lung imaging in the absence of pulmonary embolism is presented. Idiopathic pulmonary fibrosis is one of the few nonembolic diseases producing a pulmonary ventilation-perfusion mismatch. In this condition, chest radiographs may not detect the full extent of disease, and xenon-133 ventilation imaging may be relatively insensitive to morbid changes in small airways. Thus, when examining patients with idiopathic pulmonary fibrosis, one should be aware that abnormal perfusion imaging patterns without matching ventilation abnormalities are not always due to embolism. In this setting, contrast pulmonary angiography is often needed for accurate differential diagnosis.     

Pulmonary embolism

Pulmonary embolism is a common medical problem whose incidence is likely to increase in our aging population. Although it is life-threatening, effective therapy exists. The treatment is not, however, without significant complications. Thus, accurate diagnosis is important. Unfortunately, the clinical manifestations of pulmonary embolism are nonspecific. Furthermore, in many patients the symptoms of an acute embolism are superimposed on underlying chronic heart or lung disease. Thus, a high index of suspicion is needed to identify pulmonary emboli. Laboratory parameters, including arterial oxygen tensions and electrocardiography, are as nonspecific as the clinical signs. They may be more useful in excluding another process than in diagnosing pulmonary embolism. The first radiologic examination is the chest radiograph, but the clinical symptoms are frequently out of proportion to the findings on the chest films. Classic manifestations of pulmonary embolism on the chest radiograph include a wedge-shaped peripheral opacity and a segmental or lobar diminution in vascularity with prominent central arteries. However, these findings are not commonly seen and, even when present, are not specific. Even less specific findings include cardiomegaly, pulmonary infiltrate, elevation of a hemidiaphragm, and pleural effusion. Many patients with pulmonary embolism may have a normal chest radiograph. The chest radiograph is essential, however, for two purposes. First, it may identify another cause of the patient's symptoms, such as a rib fracture, dissecting aortic aneurysm, or pneumothorax. Second, a chest radiograph is essential to interpretation of the radionuclide V/Q scan. The perfusion scan accurately reflects the perfusion of the lung. However, a perfusion defect may result from a variety of etiologies. Any process such as vascular stenosis or compression by tumor may restrict blood flow. In addition, areas of the lung that are not well ventilated will be poorly perfused. Thus, a ventilation scan and a chest radiograph are essential to optimal interpretation of the perfusion scan. Ventilation/perfusion scans are interpreted as degrees of probability of pulmonary embolism. Emboli are not present in patients with a normal V/Q scan. An embolus is unlikely (10%-15%) among patients with a low-probability V/Q scan. However, small emboli that are nonocclusive may be present, and pulmonary arteriography may be used to further evaluate patients with a high clinical suspicion of pulmonary embolus.(ABSTRACT TRUNCATED AT 400 WORDS)     

Artificial neural networks (ANNs) for differential diagnosis of interstitial lung disease: results of a simulation test with actual clinical cases

RATIONALE AND OBJECTIVES: To evaluate the performance of an artificial neural network (ANN) scheme with use of consecutive clinical cases and its effect on radiologists with an observer test. MATERIALS AND METHODS: Artificial neural networks were designed to distinguish among 11 interstitial lung diseases on the basis of 26 inputs (16 radiologic findings, 10 clinical parameters). Chest radiographs of 96 consecutive cases with interstitial lung disease were used. Five radiologists independently rated their radiologic findings on the 96 chest radiographs. Based on their ratings of radiologic findings and clinical parameters obtained from the hospital information system, the output values indicating the likelihood of each of the 11 interstitial lung diseases were determined. Subsequently, 30 cases were selected from these 96 cases for an observer test. Five radiologists marked their confidence levels for diagnosis of 11 possible diseases in each case without and with ANN output. The performance of ANNs and radiologists was evaluated by receiver operating characteristic analysis based on their outputs and on confidence levels, respectively. RESULTS; The average Az value (area under the receiver operating characteristic curve) indicating ANN performance for the 96 consecutive cases was 0.85 +/- 0.03. The average Az values indicating radiologists' performance without and with ANN outputs were 0.81 +/- 0.11 and 0.87 +/- 0.06, respectively. The diagnostic accuracy was improved significantly when radiologists read chest radiographs with ANN outputs (P < .05). CONCLUSION: Artificial neural networks for differential diagnosis of interstitial lung disease may be useful in clinical situations, and radiologists may be able to utilize the ANN output to their advantage in the differential diagnosis of interstitial lung disease on chest radiographs.