Application of artificial neural networks for quantitative analysis of image data in chest radiographs for detection of interstitial lung disease

The authors have developed an automated computeraided diagnostic (CAD) scheme by using artificial neural networks (ANNs) on quantitative analysis of image data. Three separate ANNs were applied for detection of interstitial disease on digitized chest images. The first ANN was trained with horizontal profiles in regions of interest (ROIs) selected from normal and abnormal chest radiographs for distinguishing between normal and abnormal patterns. For training and testing of the second ANN, the vertical output patterns obtained from the 1st ANN were used for each ROI. The output value of the second ANN was used to distinguish between normal and abnormal ROIs with interstitial infiltrates. If the ratio of the number of abnormal ROIs to the total number of all ROIs in a chest image was greater than a specified threshold level, the image was classified as abnormal. In addition, the third ANN was applied to distinguish between normal and abnormal chest images. The combination of the rule-based method and the third ANN also was applied to the classification between normal and abnormal chest images. The performance of the ANNs was evaluated by means of receiver operating characteristic (ROC) analysis. The average Az value (area under the ROC curve) for distinguishing between normal and abnormal cases was 0.976 +/- 0.012 for 100 chest radiographs that were not used in training of ANNs. The results indicate that the ANN trained with image data can learn some statistical properties associated with interstitial infiltrates in chest radiographs.     

Computer-aided diagnosis for interstitial infiltrates in chest radiographs: optical-density dependence of texture measures

We have been developing a computerized scheme for automated detection and characterization of interstitial infiltrates based on the Fourier transform of lung texture. To improve the performance of the scheme, which was developed using digitized screen-film radiographs, optical-density dependence of both the gradient of the film used and the system noise associated with the laser scanner were investigated. Two hundred chest radiographs, including 100 abnormal cases with interstitial infiltrates, were digitized using a laser scanner. The root-mean-square (RMS) variations and the first moments of the power spectra, which correspond to the magnitude and coarseness of lung texture, were determined by Fourier transform of lung textures in numerous regions of interest (ROIs). The RMS variation was dependent upon the average optical density in the ROI, though no obvious trend existed for the first moment of the power spectrum. Dependence of the RMS variations on optical density was corrected for using the gradient curve of the film. Also, system noise associated with the laser scanner was corrected. Results indicated that the specificity was improved from 81% (without correction) to 89% (with corrections), without any loss of sensitivity (90%). Thus, the correspondence between the computer output and consensus interpretation of radiologists was improved with the new scheme compared to the previous one. This improved computerized scheme may be useful to radiologists in detecting interstitial infiltrates in chest radiographs.     

Hodgkin disease and non-Hodgkin lymphoma: plain chest radiographs and chest computed tomography of thoracic involvement in previously untreated patients

PURPOSE: To provide further information about the presentation of thoracic involvement in Hodgkin disease and non-Hodgkin lymphoma and to compare chest radiography with chest CT findings. MATERIALS AND METHODS: We reviewed the chest radiographs and the CT images of 100 Hodgkin and 100 non-Hodgkin patients, all of them untreated. Our data were compared with those of literature series: the latest study comparing the different patterns of Hodgkin and non-Hodgkin disease appeared in 1976 and it compared chest radiography with conventional tomography, not with CT. RESULTS: Intrathoracic involvement (75% vs 48%) and adenopathy (74% vs 28%) were more frequent in Hodgkin than in non-Hodgkin lymphoma. Ninety-nine per cent of the patients with intrathoracic involvement (74/75) had nodal disease. Paratracheal/prevascular nodes were most frequently involved, namely in 72/74 Hodgkin (97%) and in 27/28 non-Hodgkin patients (96%). The lung parenchyma was more often involved in non-Hodgkin (24%) than in Hodgkin (8%) patients; it was associated with mediastinal/hilar adenopathy in all Hodgkin and in 10/24 (42%) non-Hodgkin cases. Parenchymal involvement was demonstrated with chest radiography in 7/8 Hodgkin (88%) and in 13/24 non-Hodgkin patients (54%). Chest radiography showed paratracheal/prevascular adenopathy more often in Hodgkin (54/72, 75%) than in non-Hodgkin (15/27, 56%) cases. Subcarinal and internal mammary adenopathy was poorly depicted with plain films, while hilar adenopathy was generally identified with both CT and chest radiography. Chest radiography usually missed posterior mediastinal and anterior diaphragmatic adenopathy. CONCLUSIONS: The differences in the presentation of Hodgkin vs non-Hodgkin disease are not sufficiently distinctive to permit radiographic differentiation of the two conditions, but some patterns are helpful. Recognizing the frequency of thoracic involvement and that of the additional CT findings in Hodgkin and non-Hodgkin patients makes a sound basis for lymphoma imaging.     

Computerized analysis of mammographic microcalcifications in morphological and texture feature spaces

We are developing computerized feature extraction and classification methods to analyze malignant and benign microcalcifications on digitized mammograms. Morphological features that described the size, contrast, and shape of microcalcifications and their variations within a cluster were designed to characterize microcalcifications segmented from the mammographic background. Texture features were derived from the spatial gray-level dependence (SGLD) matrices constructed at multiple distances and directions from tissue regions containing microcalcifications. A genetic algorithm (GA) based feature selection technique was used to select the best feature subset from the multi-dimensional feature spaces. The GA-based method was compared to the commonly used feature selection method based on the stepwise linear discriminant analysis (LDA) procedure. Linear discriminant classifiers using the selected features as input predictor variables were formulated for the classification task. The discriminant scores output from the classifiers were analyzed by receiver operating characteristic (ROC) methodology and the classification accuracy was quantified by the area, Az, under the ROC curve. We analyzed a data set of 145 mammographic microcalcification clusters in this study. It was found that the feature subsets selected by the GA-based method are comparable to or slightly better than those selected by the stepwise LDA method. The texture features (Az = 0.84) were more effective than morphological features (Az = 0.79) in distinguishing malignant and benign microcalcifications. The highest classification accuracy (Az = 0.89) was obtained in the combined texture and morphological feature space. The improvement was statistically significant in comparison to classification in either the morphological (p = 0.002) or the texture (p = 0.04) feature space alone. The classifier using the best feature subset from the combined feature space and an appropriate decision threshold could correctly identify 35% of the benign clusters without missing a malignant cluster. When the average discriminant score from all views of the same cluster was used for classification, the Az value increased to 0.93 and the classifier could identify 50% of the benign clusters at 100% sensitivity for malignancy. Alternatively, if the minimum discriminant score from all views of the same cluster was used, the Az value would be 0.90 and a specificity of 32% would be obtained at 100% sensitivity. The results of this study indicate the potential of using combined morphological and texture features for computer-aided classification of microcalcifications.     

Ehrlichiosis: findings on chest radiographs in three pediatric patients

OBJECTIVES: Ehrlichiosis is a newly recognized tick-borne infection affecting both children and adults. The disease is usually mild, with flulike symptoms, but can be fatal. We present the chest radiographs and, in one case, the chest CT scan of all recognized pediatric cases of ehrlichiosis to date from our institution. CONCLUSION: Ehrlichiosis should be included in the differential diagnosis of acutely ill pediatric patients with a history of possible tick exposure and radiographic patterns of interstitial prominence, alveolar opacities, or consolidation that may be associated with pleural effusion.     

Automated lung segmentation in digitized posteroanterior chest radiographs

Matrix metalloproteases (MMPs) play a key role in the metabolism of connective tissue proteins in the norm and in pathology. Major MMP subfamilies (collagenases, gelatinases, and stromelysins) and matrixins, which have not been attributed to any subfamily, are reviewed. The main characteristics of these enzymes; their structural properties; their specificity; the regulation of their activity; and their role in the normal development of the matrix, the oncogenic transformation of the cell, and angiogenesis are discussed.     

Densitometric measurements of lung nodules on chest radiographs

A technique is described whereby embedded nodular densities appearing on chest radiographs can be densitometrically measured without the obscurring effects of anatomic structures. The technique consists of rear-projecting a light beam into a chest radiograph and an observe adjusting this beam until the lucency matches that of the nodule's appearance. The radiography is then removed, and the lucency is scanned and analyzed with a microcomputer. Calibration data using simulated nodules show a good accuracy in measuring size, contrast, and edge gradient.     

Mottle on computed radiographs of the chest in pediatric patients

PURPOSE: To investigate the relationship between radiation exposure and perceived mottle at bedside pediatric chest examinations performed with screen-film and computed radiographic techniques. MATERIALS AND METHODS: In a pediatric intensive care unit, chest radiographs were obtained with both computed radiography (60 radiographs) and a 600-speed screen-film system (14 radiographs). The relative radiation exposure was estimated by using the sensitivity value obtained in the processing of each computed radiograph. Five radiologists assessed the mottle present in the computed radiographs and screen-film images. RESULTS: For computed radiographs, the perceived level of mottle was inversely related to radiation exposure. For the same radiation exposure, the perceived mottle on computed radiographs was significantly higher than that on screen-film images (P < .001 for small cassettes; P < .01 for large cassettes). CONCLUSION: Pediatric computed radiography of the chest requires approximately twice the exposure of a 600-speed screen-film system to attain the same level of mottle.     

Computerized detection of masses in digitized mammograms using single-image segmentation and a multilayer topographic feature analysis

RATIONALE AND OBJECTIVES: We developed and evaluated a computer-aided detection (CAD) scheme for masses in digitized mammograms. METHODS: A multistep CAD scheme was developed and tested. The method uses a technique of single-image segmentation with Gaussian bandpass filtering to yield a high sensitivity for mass detection. A rule-based multilayer topographic feature analysis method is then used to classify suspected regions. A set of 260 cases, including 162 verified masses, was divided into two subsets; one set was used to set the rule-based classification and one was used to test the performance of the scheme. RESULTS: In a preliminary clinical study, the implemented detection scheme yielded 98% sensitivity with a false-positive detection rate of less than one false-positive region per image. CONCLUSION: Single-image segmentation methods seem to have high sensitivity in selecting true-positive mass regions in the first stage of a CAD scheme. A multilayer topographic image feature analysis method in the second stage of a CAD scheme has the potential to significantly reduce the false-positive detection rate.     

Routine portable chest radiographs in the medical intensive care unit: effects and costs

OBJECTIVE: To determine the effects and net costs of routine chest radiographs in a medical intensive care unit (ICU). DESIGN: A prospective, cohort study. A survey of experts in critical care and pulmonary diseases was undertaken to assess the effect of routine radiographs on patient management. SETTING: Medical ICU of a university hospital. PATIENTS: Eighty randomly selected patients admitted to a medical ICU. Two hundred fourteen experts were surveyed; 118 (55%)/214 responded. MEASUREMENTS AND MAIN RESULTS: Daily interviews with medical ICU clinicians were conducted to assess the radiographic findings in the routine radiographs and actions taken based on these findings. Experts evaluated the findings, their importance, the actions taken, and the probability of complications if the actions had not been taken at that time. Experts also predicted increases in length of stay associated with these complications. Presence of radiographic findings, changes in management because of the findings, net costs of routine chest radiographs, cost per finding that prompted an action, and expected changes in length of stay resulting from the actions were also assessed. Seventy-two (33%) of 221 routine radiographs (95% confidence interval: 25% to 39%) had findings, of which 44 (61%) were judged important, and 18 (8%, 95% confidence interval: 5% to 12%) prompted actions. Experts predicted that each action averted, on average, 2.1 +/- 1.7 days (SD) in the medical ICU. Mean savings per routine radiograph was $98. Net savings from routine chest radiographs remained after sensitivity analysis for expected change in length of stay, percentage of patients with routine radiographs, and percentage of routine radiographs that produce changes in management. CONCLUSION: The policy of obtaining routine chest radiographs in the medical ICU is effective and results in net savings.     

Improved parameters for unsharp mask filtering of digital chest radiographs

Observer performance with four unsharp mask filtering algorithms for storage phosphor chest radiographs was compared with that with conventional screen-film radiographs in the detection of three types of simulated lung disease: nodules, fine lines, and micronodule clusters. Previously studied parameter sets (small [1.4-mm] and medium [5-mm] filter masks) and two new parameter sets (large [2.5-cm] and ultralarge [7-cm] masks) were compared by means of receiver operating characteristic analysis. With medium and small masks, nodule detection was inferior to that achieved with other modes. Use of ultralarge masks decreased the detection of lines compared with detection with conventional screen-film radiographs. Although detection of micronodule clusters was worse with digital images than with conventional screen-film radiographs, results with large and ultralarge masks were better than with small masks. Overall, filtering with large masks was best suited for simultaneously matching linear, nodular, and micronodular structures. These results suggest that lesion-specific processing of digital chest images is not necessary.     

Detection of subtle abnormalities on chest radiographs after irreversible compression

In both oxidative phosphorylation and photo-phosphorylation, electron flow through a carrier is linked to the generation of ATP. The energy released by electron transport is converted to potential energy forming a proton gradient across the membranes in chloroplasts. The proton gradient can be measured by a pH microelectrode. In this report, pH changes produced by photo-induced proton transport through spinach chloroplast membranes were measured by a glass microelectrode. The effect of 5,5'-dithiobis(2-nitrobenzoate) (DTNB) on the kinetics of proton movement across the thylakoid membranes was studied. The results showed that the rate of proton uptake was reduced with increasing DTNB concentration. The rate of leakage of accumulated protons through thylakoid membranes also decreased. The results support the notion that cysteinyl residue is involved in proton translocation. The inhibition of proton transport would subsequently affect the chemical reactions of the Calvin Cycle that takes place in the stroma which is the soluble compartment surrounding the thylakoid membranes.     

Diagnosing left lower lobe pneumonia: usefulness of the ''spine sign'' on lateral chest radiographs

Percutaneous cardiopulmonary assist devices (PCPS) have become available in interventional cardiology within recent years. These tools offer the opportunity of performing percutaneous transluminal coronary angioplasty (PTCA) in high-risk patients characterized by significant stenoses of several coronary arteries and a poor left ventricular function. It is unclear for which patients PCPS are necessary and which patients will profit by PTCA as compared to coronary artery bypass grafting (CABG). Therefore, the anticipated risk of CABG and of PTCA without assist devices was calculated according to risk scores and compared with our results of assisted PTCA. In addition the long-term survival rate was investigated. In 35 patients (mean 65.5 years of age, 12 females, 23 males), we performed PTCA concomitant with the use of cardiac assist devices. The indications for the use of a cardiac assist device were severely impaired LV function (EF 30% +/- 8.9%) in combination with significant coronary artery disease (2.7 +/- 0.3 vessels) and a significant supply area of the vessel to be dilated. In 6 patients, PCPS was started before coronary angioplasty because of hemodynamic instability. In 21 cases, PCPS was on a standby basis without being connected to the patient's circulation. In 8 patients, a left heart assist device, the 14F-Hemopump, was inserted percutaneously. The patients were analyzed using risk scores of angioplasty and of coronary bypass graft surgery. The calculated risk of hemodynamic compromise during PTCA according to the risk scores was more than 50%. The anticipated risk of a fatal outcome following CABG would have been 19.8%. PTCA was performed on an average of 2.0 coronary arteries per patient and was successful in 85%. We observed a decline in angina pectoris classification (CCS) from 3.5 to 1.6. An average reduction of 1.1 NYHA class was achieved. The in-hospital mortality was 8.6% (3 patients: 1 x sepsis, 1 x early reocclusion, 1 x cerebral embolism). At 24 months follow-up, a re-PTCA was necessary in four cases because of restenosis. In the remainder, NYHA and CCS class were stable during the follow-up period. An additional five patients died during the first year and two patients in the second year. We conclude that PTCA with the use of a cardiac assist device shows favorable short-term results in a subset of patients with extended coronary artery disease and severely impaired LV function who are not suitable for nonsupported PTCA or CABG due to their risk profile. However, the long term results are not satisfying and stress the need for complete revascularisation with CABG once the patient's condition is stabilized by means of supported PTCA.     

Computerized radiographic analysis of osteoporosis: preliminary evaluation

Measurement of bone mass is important in determining the risk for fracture and in following the course of patients undergoing therapy for osteoporosis. Bone mineral densitometry (BMD) is a good predictor of fracture risk, but there is considerable overlap in BMD measurements between individuals with fracture risk and those without. In this study, computerized texture analysis of the trabecular pattern on conventional spine radiographs was used to evaluate bone structure as a determinant of fracture risk. Standard lumbar spine radiographs of 43 individuals were analyzed and compared with BMD measurements obtained with dual-photon absorptiometry. This method was more effective than BMD in differentiation of patients with fractures elsewhere in the spine from those with no fracture. These preliminary results suggest that this method of bone structure analysis, combined with BMD, may lead to a more sensitive and specific predictor of osteoporosis and risk of fracture.     

Computerized axial transverse tomography in cerebrovascular disease

One hundred eleven patients with supratentorial cerebrovascular disease were studied by computerized axial tomography (CT scanning). With one exception, every patient who had a normal scan 48 hours after the onset of symptoms was ultimately diagnosed as having had transient ischemic attack, although in nearly one-third, the clinical diagnoses at the time of the scan was infarction. A normal CT scan, therefore, augurs a good outcome of supratentorial cerebrovascular disease. Ninety-eight percent of the patients with infarction had abnormal scans, with areas of decreased density in a vascular distribution. Pitfalls in the diagnosis of infarction were (1) initially normal CT scans that changed to abnormal after 48 hours, and (2) mass effect of infarction leading to misdiagnosis of brain tumor. Serial studies eliminated both pitfalls. Intracerebral hemorrhages had a distinctive high density appearance. In 43 percent of patients whose scans showed hemorrhage, the clinical diagnosis was thrombosis. Many did not have symptoms, signs, or outcome of cerebral hemorrhage, and the diagnosis would not have been suspect were it not for the CT scan.     

Morphological changes in chest radiographs of patients with acute respiratory distress syndrome (ARDS)

The value of ultrasonography (US), magnetic resonance imaging (MRI), computed tomography (CT) and plain radiography (PR) in detecting bone erosions on the humeral head was evaluated in a study of 26 in-patients (26 shoulders) with rheumatoid arthritis (RA). MRI depicted humeral erosions in 25 (96%), US in 24 (92%), CT in 20 (77%) and PR in 19 (73%) of the 26 shoulders. MRI and US were superior to CT in detecting small erosions. US was the most sensitive method to show surface erosions on the greater tuberosity. US, CT and MRI detected large erosions quite similarly. PR frequently missed small erosions. In the evaluation of early erosions in the rheumatoid shoulder, US and MRI are more sensitive methods than the traditionally used PR. US and MRI are suitable for the evaluation of soft-tissue involvement in the rheumatoid shoulder, but also for the detection of bone erosions of the humeral head.     

Pulmonary parenchymal manifestations surrounding small peripheral masses: pathologic correlation with chest radiographs and diagnostic value

The corrosion of aluminium (Al) in several brands of soft drinks (cola- and citrate-based drinks) has been studied, using an electrochemical method, namely potentiodynamic polarization. The results show that the corrosion of Al in soft drinks is a very slow, time-dependent and complex process, strongly influenced by the passivation, complexation and adsorption processes. The corrosion of Al in these drinks occurs principally due to the presence of acids: citric acid in citrate-based drinks and orthophosphoric acid in cola-based drinks. The corrosion rate of Al rose with an increase in the acidity of soft drinks, i.e. with increase of the content of total acids. The corrosion rates are much higher in the cola-based drinks than those in citrate-based drinks, due to the facts that: (1) orthophosphoric acid is more corrosive to Al than is citric acid, (2) a quite different passive oxide layer (with different properties) is formed on Al, depending on whether the drink is cola or citrate based. The method of potentiodynamic polarization was shown as being very suitable for the study of corrosion of Al in soft drinks, especially if it is combined with some non-electrochemical method, e.g. graphite furnace atomic absorption spectrometry (GFAAS).     

Anatomic region-based dynamic range compression for chest radiographs using warping transformation of correlated distribution

SCO-spondin is a newly identified protein, strongly expressed in the subcommissural organ (SCO), an ependymal differentiation of the brain. When secreted into the cerebrospinal fluid at the entrance to the Sylvian aqueduct, it condenses and forms Reissner's fiber. Several conserved domains have previously been characterizedin SCO-spondin, e.g., thrombospondin type 1 repeats (TSRs), low-density lipoprotein receptor (LDLr) type A repeats, and epidermal-growth-factor-like domains, which are potent sites of protein-protein interaction. To clarify the role of this protein on neuronal development, we have tested the effect of oligopeptides, the sequences of which include highly conserved amino acids of TSRs, LDLr type A repeats and a potent site of attachment to proteoglycan, on cortical and spinal-cord neurons in primary cell cultures. One of these peptides (WSGWSSCSRSCG), corresponding to a SCO-spondin TSR sequence, markedly increases adhesivity and neuritic outgrowth of cortical neurons and induces an opposite effect on cortical and spinal-cord neuronal aggregation. These effects are specific, as no response is observed with the scrambled sequence of this peptide. Another peptide (WGPCSVSCG) is only slightly active on adhesivity and neuritic outgrowth of cortical neurons and has no effect on spinal-cord neurons. Peptides derived from other conserved domains of SCO-spondin are not effective under our experimental conditions. Thus, SCO-spondin may be responsible for at least a part of the effects previously observed on neuronal cells cultured in the presence of Reissner's fiber. In addition, SCO-spondin seems to interfere with neuronal development and/or axonal guidance during ontogenesis of the central nervous system in modulating side-to-side interactions and neuritic outgrowth.